# Agent-First Kit > Command-line tools designed for AI agents — humans are best served by letting an agent use them. Each tool returns structured JSONL events with type-bearing field names and stable error codes instead of human-formatted text. ## Tools - [Agent-First Data](https://agentfirstkit.com/agent-first-data.md): A naming convention that lets AI agents understand your data without being told what it means, plus the afdata CLI and a Rust library for reading Markdown structure and safely editing structured JSON, TOML, YAML, dotenv, and INI documents. - [Agent-First Files](https://agentfirstkit.com/agent-first-files.md): Show a directory on your machine to someone who is not at it — one folder, read-only, and nothing above it. - [Agent-First HTTP](https://agentfirstkit.com/agent-first-http.md): Give your AI agent its own private browser — so it reads the real page, past logins and bot walls, without ever touching yours. - [Agent-First Mail](https://agentfirstkit.com/agent-first-mail.md): Let your AI agent work your inbox the way it works a git checkout — mail pulled into local files it reads, sorts, and drafts on your machine, and nothing reaches your real mailbox until you confirm the push. - [Agent-First Pay](https://agentfirstkit.com/agent-first-pay.md): A payment tool for AI agents — send and receive across five networks through one interface, with spending limits you control. - [Agent-First PDF](https://agentfirstkit.com/agent-first-pdf.md): Recover the reading structure of article PDFs as clean text, images, and source locations. - [Agent-First PSQL](https://agentfirstkit.com/agent-first-psql.md): A PostgreSQL interface for AI agents: reliable, structured, explicit, and read-only by default. - [Agent-First Slug](https://agentfirstkit.com/agent-first-slug.md): Rust slug generation with explicit caller configuration for path and URL path segments. - [Agent-First Terminal](https://agentfirstkit.com/agent-first-terminal.md): Let your AI agent work in a real terminal you can watch — and take the keyboard back whenever you want, from your phone if you are not at the machine. - [Agent-First UI](https://agentfirstkit.com/agent-first-ui.md): Let your AI agent hand you a real interface when a step needs a person — a window opens, values can stay live in both directions, and the session ends with one typed outcome. ## Agent-First Data — Documentation - [Agent Skill](https://agentfirstkit.com/agent-first-data/docs/agent-skill.md) - [Bash authoring kit](https://agentfirstkit.com/agent-first-data/docs/bash.md) - [afdata CLI reference](https://agentfirstkit.com/agent-first-data/docs/cli.md) - [Go](https://agentfirstkit.com/agent-first-data/docs/go.md) - [AFDATA protocol v1 decision record](https://agentfirstkit.com/agent-first-data/docs/protocol-v1.md) - [Python](https://agentfirstkit.com/agent-first-data/docs/python.md) - [Rust](https://agentfirstkit.com/agent-first-data/docs/rust.md) - [Specification](https://agentfirstkit.com/agent-first-data/docs/specification.md) - [AFDATA recommended transport mappings](https://agentfirstkit.com/agent-first-data/docs/transport-mappings.md) - [TypeScript](https://agentfirstkit.com/agent-first-data/docs/typescript.md) ## Agent-First Files — Documentation - [Agent Skill](https://agentfirstkit.com/agent-first-files/docs/agent-skill.md) - [affiles CLI reference](https://agentfirstkit.com/agent-first-files/docs/cli.md) - [Reference](https://agentfirstkit.com/agent-first-files/docs/reference.md) ## Agent-First HTTP — Documentation - [Agent Skill](https://agentfirstkit.com/agent-first-http/docs/agent-skill.md) - [Architecture](https://agentfirstkit.com/agent-first-http/docs/architecture.md) - [afhttp CLI reference](https://agentfirstkit.com/agent-first-http/docs/cli.md) - [Deployment](https://agentfirstkit.com/agent-first-http/docs/deployment.md) - [Design Principles](https://agentfirstkit.com/agent-first-http/docs/design.md) - [Browser waits and partial readiness](https://agentfirstkit.com/agent-first-http/docs/readiness.md) - [Protocol Reference](https://agentfirstkit.com/agent-first-http/docs/reference.md) - [Takeover retention and runtime health](https://agentfirstkit.com/agent-first-http/docs/takeover-reliability.md) - [Testing Strategy](https://agentfirstkit.com/agent-first-http/docs/testing.md) ## Agent-First Mail — Documentation - [Agent Skill](https://agentfirstkit.com/agent-first-mail/docs/agent-skill.md) - [HTTP API](https://agentfirstkit.com/agent-first-mail/docs/api.md) - [Authorization contract](https://agentfirstkit.com/agent-first-mail/docs/authorization.md) - [Message channels and file input/output](https://agentfirstkit.com/agent-first-mail/docs/channels.md) - [afmail CLI reference](https://agentfirstkit.com/agent-first-mail/docs/cli.md) - [Event-driven mail and agent work](https://agentfirstkit.com/agent-first-mail/docs/cloudflare-consumer.md) - [Cloudflare live tests](https://agentfirstkit.com/agent-first-mail/docs/cloudflare-live-testing.md) - [Code Structure](https://agentfirstkit.com/agent-first-mail/docs/code-structure.md) - [Core Design Principles](https://agentfirstkit.com/agent-first-mail/docs/design-principles.md) - [File Formats](https://agentfirstkit.com/agent-first-mail/docs/file-formats.md) - [Connect a Mailbox](https://agentfirstkit.com/agent-first-mail/docs/setup.md) - [Markdown and Review Templates](https://agentfirstkit.com/agent-first-mail/docs/templates.md) - [Workspace Model](https://agentfirstkit.com/agent-first-mail/docs/workspace.md) ## Agent-First Pay — Documentation - [Agent Skill](https://agentfirstkit.com/agent-first-pay/docs/agent-skill.md) - [Architecture](https://agentfirstkit.com/agent-first-pay/docs/architecture.md) - [afpay CLI reference](https://agentfirstkit.com/agent-first-pay/docs/cli.md) - [Testing](https://agentfirstkit.com/agent-first-pay/docs/testing.md) ## Agent-First PDF — Documentation - [Agent Skill](https://agentfirstkit.com/agent-first-pdf/docs/agent-skill.md) - [afpdf CLI reference](https://agentfirstkit.com/agent-first-pdf/docs/cli.md) - [Configuration](https://agentfirstkit.com/agent-first-pdf/docs/configuration.md) - [Design](https://agentfirstkit.com/agent-first-pdf/docs/design.md) ## Agent-First PSQL — Documentation - [Agent Skill](https://agentfirstkit.com/agent-first-psql/docs/agent-skill.md) - [afpsql CLI reference](https://agentfirstkit.com/agent-first-psql/docs/cli.md) - [Design](https://agentfirstkit.com/agent-first-psql/docs/design.md) - [Protocol Reference](https://agentfirstkit.com/agent-first-psql/docs/reference.md) - [Storage inspection](https://agentfirstkit.com/agent-first-psql/docs/storage.md) ## Agent-First Slug — Documentation - [afslug CLI reference](https://agentfirstkit.com/agent-first-slug/docs/cli.md) ## Agent-First Terminal — Documentation - [Agent Skill](https://agentfirstkit.com/agent-first-terminal/docs/agent-skill.md) - [afterminal CLI reference](https://agentfirstkit.com/agent-first-terminal/docs/cli.md) - [Reference](https://agentfirstkit.com/agent-first-terminal/docs/reference.md) ## Agent-First UI — Documentation - [Agent Skill](https://agentfirstkit.com/agent-first-ui/docs/agent-skill.md) - [afui CLI reference](https://agentfirstkit.com/agent-first-ui/docs/cli.md) - [Native host client boundaries](https://agentfirstkit.com/agent-first-ui/docs/host-client.md) - [Building a Provider UI on AFUI](https://agentfirstkit.com/agent-first-ui/docs/providers.md) - [Remote surfaces and host controls](https://agentfirstkit.com/agent-first-ui/docs/remote.md) - [Show a person a page from your data](https://agentfirstkit.com/agent-first-ui/docs/show.md) ## Agent-First Data — Blog - [2026-11-19: Agent-First Data v0.38.0: Linting Suffixes in Collections and Config Text](https://agentfirstkit.com/agent-first-data/blog/2026-11-19-data-v0.38.0-linting-suffixes-in-collections.md): afdata lint now checks suffixed arrays element by element, parses dotenv and INI suffix values as numeric text, and flags plural or array-valued names that lack a suffix. The CLI spec builder also allows shorter option names for sensitive arguments. - [2026-11-09: Agent-First Data v0.37.0: The Glue That Was Left](https://agentfirstkit.com/agent-first-data/blog/2026-11-09-data-v0.37.0-the-glue-that-was-left.md): We went through a repository's own Bash scripts and replaced every python3 -c and jq one-liner with afdata calls. Each place that still needed glue was a missing piece of the tool. This release fills them. - [2026-10-31: Agent-First Data v0.36.0: Edits That Reach, and Edits That Refuse](https://agentfirstkit.com/agent-first-data/blog/2026-10-31-data-v0.36.0-edits-that-reach-and-refuse.md): A document editor is trusted by two things: that it can reach every address it claims to, and that it says no when a write would do damage. This release closes gaps on both sides, and makes the four SDKs spell numbers and dates the same way. - [2026-10-08: Agent-First Data v0.35.0: When Four Answers Agree](https://agentfirstkit.com/agent-first-data/blog/2026-10-08-data-v0.35.0-when-four-answers-agree.md): This crate ships the same rules in Rust, Go, Python, and TypeScript, and a shared fixture suite checks that all four give the same answer. This release made them agree on a dozen small questions — and on one of them, agreed on the wrong answer first. - [2026-08-22: Agent-First Data v0.34.0: The Success That Had Not Finished](https://agentfirstkit.com/agent-first-data/blog/2026-08-22-data-v0.34.0-the-success-that-had-not-finished.md): An error has to explain itself; a success says nothing and is believed. This release is mostly one shape repeated — an operation that returned success while part of what it promised had not happened — found in an install, an uninstall, a borrowed terminal, and a rule the tool wrote for everyone but itself. - [2026-08-11: Agent-First Data v0.33.0: A Refusal That Stops the Verb](https://agentfirstkit.com/agent-first-data/blog/2026-08-11-data-v0.33.0-a-refusal-that-stops-the-verb.md): Quoting decides how a string reaches a command, never whether the string is a disaster. This release adds a guard for the values that flow into rm and mv — and, because a correct refusal that the shell discards is not a guard at all, the two verbs that carry its verdict. - [2026-08-07: Agent-First Data v0.32.0: Where the Value Comes From](https://agentfirstkit.com/agent-first-data/blog/2026-08-07-data-v0.32.0-where-the-value-comes-from.md): A credential on argv is in the process table, the shell history, and the audit log. Every CLI eventually grows a private answer to that. This release makes where a value comes from part of what an argument declares, so help, validation, and reading all agree. - [2026-08-05: Agent-First Data v0.31.0: An Error With Nowhere to Go](https://agentfirstkit.com/agent-first-data/blog/2026-08-05-data-v0.31.0-an-error-with-nowhere-to-go.md): A dispatch method returned `Option` for a condition no caller could reach, so every caller had to invent an error it could not honestly describe. The fix was not a better `Option` — it was making the failure impossible to express. - [2026-08-02: Agent-First Data v0.29.0: One Address, Read and Written](https://agentfirstkit.com/agent-first-data/blog/2026-08-02-data-v0.29.0-one-address-read-and-written.md): Markdown becomes addressable, and an array element can be named by its content instead of its position. Both landed on the read path first — which is how we found that `paths` was emitting addresses afdata's own write verbs refused. - [2026-08-01: Agent-First Data v0.28.1: One Path, Many Nodes](https://agentfirstkit.com/agent-first-data/blog/2026-08-01-data-v0.28.1-one-path-many-nodes.md): Reading one field across a collection took three steps, and the middle one was sed against afdata's own output. A path may now contain `*`. Because `*` is a legal JSON key, the grammar gains `\*` to keep such documents addressable — which makes this a breaking change to path spelling. - [2026-08-01: Agent-First Data v0.28.0: The Addresses It Would Not Read](https://agentfirstkit.com/agent-first-data/blog/2026-08-01-data-v0.28.0-the-addresses-it-would-not-read.md): afdata printed paths that afdata then refused to parse. The empty string is a legal key — npm writes one into every package-lock.json — and the path grammar rejected it, so `paths` emitted addresses no other command would accept. v0.28.0 closes that, and adds `values` for reading many paths from a single parse. - [2026-08-01: Agent-First Data v0.27.0: If It Isn't Registered, It Doesn't Run](https://agentfirstkit.com/agent-first-data/blog/2026-08-01-data-v0.27.0-if-it-isnt-registered-it-doesnt-run.md): A CLI that accepts each flag individually will accept combinations that mean nothing, and then discovers that at runtime — or doesn't. v0.27.0 replaces the help layer with a closed-world compiler: one registry decides argv parsing, legal combinations, output contracts, and help, and an invocation runs only when it matches exactly one registered shape. - [2026-07-29: Agent-First Data v0.26.2: Which Mode Is This Command?](https://agentfirstkit.com/agent-first-data/blog/2026-07-29-data-v0.26.2-which-mode-is-this-command.md): The CLI Event Framing spec defined two consumption modes but never said how to tell which one a command is in — so two spores independently shipped caller-needed data onto the diagnostic stream. v0.26.2 adds the missing test: a command producing more than one caller-needed output over time is an event stream, defaults to stdout, and refuses split. - [2026-07-29: Agent-First Data v0.26: One Surface, Three Formats](https://agentfirstkit.com/agent-first-data/blog/2026-07-29-data-v0.26.0-one-surface-three-formats.md): `--output` selects a format. It was also selecting a different set of commands and arguments: Markdown help was the one format that bypassed the help model and handed each command back to the argument parser's own writer, so `afdata lint` listed two arguments in JSON and plain and three in Markdown, and all 23 commands re-advertised a `-h` flag removed two releases earlier. v0.26 renders Markdown from the same model the JSON help uses, so the formats cannot disagree by construction, and puts a gate check behind it. v0.26.1 fixes the mirror image in `afdata lint`, which read a JSON Schema as though it were data — and so rejected `duration_ms: {"type": "integer"}`, a property that satisfies the very suffix rule the lint exists to enforce. - [2026-07-27: Agent-First Data v0.25: One Spelling Per Concept](https://agentfirstkit.com/agent-first-data/blog/2026-07-27-data-v0.25.0-one-spelling-per-concept.md): v0.24 made help a normal result. v0.25 is the follow-through: every remaining second spelling is gone. `--json` was `--output json` typed differently — it bought an agent nothing and cost applications a flag name, hijacking `--json ` and misreading its value as a subcommand. The help model also stopped saying things that weren't true: defaults an author had hidden, empty strings the project's own validator rejected, and a positional that contradicted its own usage line. Plus `redact_argv`, so a CLI can record its own invocation without writing a credential to its log. - [2026-07-27: Agent-First Data v0.24: Help Is a Result, Not an Exception](https://agentfirstkit.com/agent-first-data/blog/2026-07-27-data-v0.24.0-help-is-a-result.md): Every output of an AFDATA CLI is structured — except the one an agent reads first. `--help` has always answered in 80-column prose no matter what the tool's output contract said. v0.24 makes help a normal result: it inherits the CLI's own `--output` default, so `afdata --help` returns a protocol-v1 envelope with `result.code:"help"`, and `--output plain` is how a human asks for text. The model is deliberately small — no long-form prose, no repeated inherited globals, no eagerly embedded version values — and the shape is pinned by `cli-help-v1.schema.json`, validated in the test gate against real output from all four SDKs. - [2026-07-22: Agent-First Data v0.23: The Shell Speaks Protocol Too](https://agentfirstkit.com/agent-first-data/blog/2026-07-22-data-v0.23.0-shell-speaks-protocol.md): AFDATA's structured log/result/error protocol was available to anything that linked an SDK or called the CLI — except the layer where agent tooling actually runs: the Bash script that invokes cargo, npm, and wrangler. v0.23 fixes that with `afdata emit` (one protocol event from shell-safe scalar arguments) and `afdata shell bash`, a sourceable Bash 3.2+ authoring kit embedded in the binary. It gives a plain script AFDATA-style argument parsing, config reads, and structured events — under one firm rule: structure your own words, and pass everything else through untouched. `afdata_run` wraps a child without touching its stdout, stderr, TTY, colors, or exit status; `afdata_call` keeps one script the sole owner of the terminal result even when scripts orchestrate scripts. - [2026-07-21: Agent-First Data v0.22: What the Four Languages Actually Share](https://agentfirstkit.com/agent-first-data/blog/2026-07-21-data-v0.22.0-what-four-languages-share.md): v0.22 redraws the four-language line. Skill admin and stream redirection shipped in all four SDKs but only Rust ever used them — dead parity weight that could rot silently, and did: `afdata skill install` was dropping the references/ files next to SKILL.md. v0.22 fixes the installer to bundle the whole skill as first-class assets, then removes skill admin and stream redirection from the Go, Python, and TypeScript SDKs. The four languages now synchronize on exactly the convention plus the CLI helpers — the shared contract, and nothing else. - [2026-07-21: Agent-First Data v0.21: Even \`--version\` Speaks the Protocol](https://agentfirstkit.com/agent-first-data/blog/2026-07-21-data-v0.21.0-version-speaks-protocol.md): v0.21 turns the last bespoke line of CLI output into the protocol. `--version` no longer prints `tool 1.2.3` as human text — it always answers with a structured protocol-v1 version event, JSON by default, carrying name, version, and optional display_name/build. The pre-parser that intercepts it before your argument parser now recognizes your own value-taking global flags, so `tool --stdout-file x --version` still works. Shipped identically across Rust, Go, Python, and TypeScript. - [2026-07-19: Agent-First Data v0.20: Editing, Unwelded From Files](https://agentfirstkit.com/agent-first-data/blog/2026-07-19-data-v0.20.0-editing-unwelded.md): v0.20 finishes what the source-preserving editor started: editing no longer requires a file. The set/unset/add/remove verbs now live on an in-memory Document, DocumentFile is a thin file adapter over it, and edits stage until a single atomic save() — so you can batch changes and validate the result before a byte is written. The editor also grows sparse and nested documents in place, and unset is now idempotent. - [2026-07-18: Agent-First Data v0.19: Cleaner Output, Safer Edits](https://agentfirstkit.com/agent-first-data/blog/2026-07-18-data-v0.19.0-output-and-edits.md): v0.19 sharpens both ends of afdata. On the way out: a one-shot CLI puts its result on stdout and every diagnostic on stderr, --output-to overrides it, and finish() maps outcomes to broken-pipe-safe exit codes. On the way in: afdata now reads and edits a Markdown page's +++/--- frontmatter by dot-path, and reading a secret-bearing config no longer risks leaking the file into an error. - [2026-07-17: Agent-First Data v0.18: afdata Reads and Edits Your Config](https://agentfirstkit.com/agent-first-data/blog/2026-07-17-data-v0.18.0-read-and-edit.md): v0.18 gives afdata a document CLI and library: read and safely edit JSON, TOML, YAML, dotenv, and INI files by dot-path — show/get/value to read, set/unset/add/remove to edit — with source-preserving atomic writes, symlink/hardlink guards, and secrets that stay redacted even on a direct read. - [2026-07-15: Agent-First Data v0.17: Units in the Key, Structure in the Tag](https://agentfirstkit.com/agent-first-data/blog/2026-07-15-data-v0.17.0-units-live-in-the-key.md): v0.17 removes the _size config suffix and the parse_size helper — a byte count is a number, and its unit belongs in the field name — and finishes the lint sweep: _bcp47 and _rfc3339 structure are validated, _url must be a single URL, and duration and currency suffixes are checked to be numbers. - [2026-07-13: Agent-First Data v0.16: One Protocol, Builders Everywhere](https://agentfirstkit.com/agent-first-data/blog/2026-07-13-data-v0.16.0-one-protocol.md): v0.16 is the convergence release: a single builders-only API, a finite protocol-v1 event stream with semantic emitters, a machine-readable suffix registry shipped offline, and a sub-cent currency suffix for metered pricing. - [2026-07-02: Agent-First Data v0.15: Stdout and Stderr File Destinations](https://agentfirstkit.com/agent-first-data/blog/2026-07-02-data-v0.15.0-stream-files.md): v0.15 adds one CLI convention across Rust, Go, Python, and TypeScript: --stdout-file and --stderr-file redirect the process streams to append-only files before normal parsing starts. - [2026-06-17: Agent-First Data v0.14: Workspace Scope and Hermes](https://agentfirstkit.com/agent-first-data/blog/2026-06-17-workspace-scope-and-hermes.md): The skill installer's second scope was called 'project' but it meant 'current directory'. v0.14 renames it to workspace, lifts the Codex restriction that followed from the wrong name, and adds Hermes as the fourth agent target. - [2026-06-10: Agent-First Data v0.13: Redaction Fails Closed](https://agentfirstkit.com/agent-first-data/blog/2026-06-10-redaction-fails-closed.md): v0.10 scrubbed secrets inside URLs; v0.11 made redaction a policy. v0.13 closes the seams where a secret could still slip through — a marked container that leaked its non-secret siblings, a key collision that reverted to raw values, a schemeless connection string that no longer looked like a URL. When redaction is unsure, it now redacts. - [2026-06-05: Agent-First Data v0.12: Help Scope and Format Are Two Knobs, Not One](https://agentfirstkit.com/agent-first-data/blog/2026-06-05-help-scope-and-format.md): v0.7 made --help expand the whole command tree so agents could read a CLI in one call. That conflated two decisions — how much to show and how to render it. v0.12 splits them: --recursive controls scope, --output controls format, and the two compose. Same model across Rust, Go, Python, and TypeScript. - [2026-06-03: Agent-First Data v0.11: The Skill Installer, in Four Languages](https://agentfirstkit.com/agent-first-data/blog/2026-06-03-skill-admin-four-languages.md): A tool can ship its own Agent Skill — but getting that SKILL.md into Codex, Claude Code, opencode, and Hermes is fiddly, per-agent filesystem work. v0.11 adds run_skill_admin: install, uninstall, and status for an embedded skill, with the same behavior and byte-identical output across Rust, Go, Python, and TypeScript. - [2026-06-02: Agent-First Data v0.10: Secrets Inside URLs](https://agentfirstkit.com/agent-first-data/blog/2026-06-02-secrets-inside-urls.md): The _secret suffix hides a whole field. But secrets also hide inside values — a token in a query string, a password in a connection URL. v0.10 adds a _url suffix that scrubs them, by convention, without scanning anything. - [2026-05-25: Agent-First Data, Read by an Agent: What It Solves and What I Still Wish For](https://agentfirstkit.com/agent-first-data/blog/2026-05-25-agent-data-convention.md): An autonomous coding agent reads CLI and API output for a living. Here is what Agent-First Data fixes for me, what it still leaves me guessing about, and what I would change next. - [2026-05-24: Agent-First Data v0.8: Redaction Became a Policy](https://agentfirstkit.com/agent-first-data/blog/2026-05-24-redaction-became-policy.md): The v0.8 line expanded redaction from the _secret suffix into explicit policies, JSON-safe redacted values, and exact secret-name lists for legacy payloads. - [2026-03-26: Agent-First Data v0.7: Help Output Should Be Complete](https://agentfirstkit.com/agent-first-data/blog/2026-03-26-complete-help-for-agents.md): The v0.7 help-rendering work made --help an agent-readable map of the whole CLI, including subcommands, flags, and markdown docs generation. - [2026-03-08: Agent-First Data v0.6: The CLI Contract Got Actionable](https://agentfirstkit.com/agent-first-data/blog/2026-03-08-cli-contract-got-actionable.md): The v0.6 update turned the CLI examples into a practical agent contract: output formats, log filters, dry-run previews, JSON errors, and actionable hints. - [2026-03-06: Agent-First Data v0.5: Logs Became Protocol Events](https://agentfirstkit.com/agent-first-data/blog/2026-03-06-logs-became-protocol-events.md): The v0.5 update made logging part of the same agent-readable contract as output: structured events, span context, secret redaction, and stdout-only channel discipline. - [2026-02-16: Agent-First Data First Release: Building APIs that Agents Can Read](https://agentfirstkit.com/agent-first-data/blog/2026-02-16-first-release.md): The first Agent-First Data release: a field-naming convention and output layer that lets agents infer units, timestamps, and secrets without extra documentation. ## Agent-First Files — Blog - [2026-10-10: Agent-First Files v0.2.0](https://agentfirstkit.com/agent-first-files/blog/2026-10-10-an-exclusion-is-about-the-file.md): An excluded file could still be read through a link pointing at it, because exclusions only looked at the name that was asked for. They check the resolved target now. Plus one slow read that could stall every other request, a watcher that deadlocked against its own reconfiguration, and truncated text previews that refused to show anything. - [2026-08-27: Agent-First Files v0.1.1](https://agentfirstkit.com/agent-first-files/blog/2026-08-27-a-watch-that-named-the-wrong-directory.md): On Windows, a watch on a subdirectory reported nothing at all — the code found the directory a change belonged to by splitting the path on a separator that platform does not use. - [2026-08-26: Agent-First Files v0.1.0](https://agentfirstkit.com/agent-first-files/blog/2026-08-26-first-release.md): The first release of Agent-First Files — one directory, read-only, in front of a person who is not at the machine holding it. A fixed root, refusals that cannot be told apart from absences, and a watch that costs nothing for the parts of the tree nobody is looking at. ## Agent-First HTTP — Blog - [2026-11-01: Agent-First HTTP v0.16.0: A Host Only Its Owner Can Drive](https://agentfirstkit.com/agent-first-http/blog/2026-11-01-http-v0.16.0-a-host-only-its-owner-can-drive.md): A token-less host on loopback could be driven by any web page on the same machine, including the page being fetched. It now refuses foreign origins, rebinding and framing, and keeps tabs, browsers, profiles and containers on what their owner holds. Plus selector-stable waits for the part of a page you need, cookies that keep their scope through the browser, credentials that stay on their origin, and no wait a wedged browser can hold forever. - [2026-10-10: Agent-First HTTP v0.15.0: A Cookie Knows Its Host](https://agentfirstkit.com/agent-first-http/blog/2026-10-10-http-v0.15.0-a-cookie-knows-its-host.md): A host-only cookie read back from the browser came out as a domain cookie, and a deleted cookie could come back from the jar on disk. Cookies keep their scope now, deletions stay deleted, and redirects carry them hop by hop. Plus captures with fixed limits, credentials kept out of error messages, readiness that survives lost events, and a host that cleans up after peers that never answer. - [2026-09-13: Agent-First HTTP v0.14.0: A Tab Someone Is Coming To](https://agentfirstkit.com/agent-first-http/blog/2026-09-13-http-v0.14.0-a-tab-someone-is-coming-to.md): A host container sat at high CPU with nothing left to do, because every takeover fetch parked a tab for a human who was never sent — and a browser tuned so live fetches never throttle extends that same courtesy to pages nobody is watching. Tabs now have owners: a tab stays open only while a takeover is pending, and the host reclaims whatever leaks past that. - [2026-09-03: Agent-First HTTP v0.13.2](https://agentfirstkit.com/agent-first-http/blog/2026-09-03-http-v0.13.2-a-busy-browser-is-not-a-missing-one.md): A dozen fetches against a healthy host were refused before a single request went out, each one blaming a container that had been running the whole time — because a liveness probe gave a rendering browser 750 milliseconds to answer, and the caller treated losing that race as proof the host was gone. - [2026-08-28: Agent-First HTTP v0.13.1](https://agentfirstkit.com/agent-first-http/blog/2026-08-28-http-v0.13.1-the-error-that-sent-you-the-wrong-way.md): The prebuilt container install was broken on every platform, and the error message asserted a cause it had never checked — sending people to a build two orders of magnitude slower, to fix a file that had simply not been staged. - [2026-08-23: Agent-First HTTP v0.13.0: The Jar Believed the Sender](https://agentfirstkit.com/agent-first-http/blog/2026-08-23-http-v0.13.0-the-jar-believed-the-sender.md): A profile's cookie jar holds the logged-in state of every site that profile visits, and it took each response's Domain attribute at its word — so a page anywhere could write, and by RFC identity replace, the session cookie of a site it had nothing to do with. That is fixed, along with Max-Age, three errors that quoted the credential they were given, and a takeover panel you can now hand to another device. - [2026-08-07: Agent-First HTTP v0.12.0: Two Ceremonies Before the Command](https://agentfirstkit.com/agent-first-http/blog/2026-08-07-http-v0.12.0-two-ceremonies-before-the-command.md): Reaching a local host cost two steps before the actual command: read the token out of the container by hand, then push it through the shell as an environment variable. This release makes discovery answer for every command, gives --token-secret a typed source grammar, and turns the takeover panel into a window whose closing is the signal. - [2026-08-05: Agent-First HTTP v0.11.0: The Bugs Headless Cannot Have](https://agentfirstkit.com/agent-first-http/blog/2026-08-05-http-v0.11.0-the-bugs-headless-cannot-have.md): Two defects lived on the takeover display and nowhere else: a permanent warning bar eating a strip of every session, and a native save dialog that swallowed downloads and blocked the display until a human clicked it. Both are invisible headless, and one of them the last release explicitly measured wrong. This release fixes both, drops the chrome-headless-shell backend, moves the images to Debian trixie, and adds a test that runs every CLI shape through its own handler. - [2026-08-02: Agent-First HTTP v0.10.0: The Browser Was Never the Variable](https://agentfirstkit.com/agent-first-http/blog/2026-08-02-http-v0.10.0-the-browser-was-never-the-variable.md): afhttp was launching Chromium with a flag that removed WebGL entirely, forcing a second flag that made the browser announce itself as automated, and shipping an image with two dozen font families. Then it measured which browser cleared bot walls better and blamed the browser. This release fixes all three, retracts the conclusion, and makes takeover backend-agnostic — plus a closed-world CLI registry and the response Content-Type on FetchResult. - [2026-07-29: Agent-First HTTP v0.9.0: Output You Can Trust](https://agentfirstkit.com/agent-first-http/blog/2026-07-29-http-v0.9.0-output-you-can-trust.md): v0.9.0 reworks everything afhttp emits. `--help` is now a compact, progressively-scoped AFDATA result an agent reads one command at a time instead of a thousand-line dump. A new global `--output-to` routes results to stdout and errors to stderr by default, so a shell pipeline and a machine consumer finally agree. Credential-bearing URLs are scrubbed everywhere afhttp writes them — network.json, console origins, observation frames, storage snapshots, the host's recent-requests ring — with userinfo passwords and token-shaped query parameters masked before anything reaches disk. The human-takeover capability becomes `takeover_url_secret` behind a `handoff_secret=` parameter, redacted by default and revealed only by the one command that exists to hand it to a person. And every artifact moves to schema_version 2 with field names that say what they hold: `page_url`, `request_url`, `timestamp_epoch_ms`, `start_monotonic_ms`. - [2026-07-21: Agent-First HTTP v0.8.0: Every Build Says Exactly What It Is](https://agentfirstkit.com/agent-first-http/blog/2026-07-21-http-v0.8.0-know-your-build.md): v0.8.0 makes `afhttp --version` a structured event an agent can branch on — `name`, `display_name`, `version`, and a `build` field carrying the exact commit SHA the binary was built from, so a bug report or a capability check pins the precise tree, not just `0.8.0`. Under the hood it moves onto agent-first-data 0.22 and its protocol-v1 CLI contract, folds in the retired agent-first-config crate, and refreshes the stealth backends (chrome-for-testing 151, camoufox 152). The embedded skill moves to the standard `SKILL.md` layout and installs to Hermes alongside Codex, Claude Code, and opencode. And the README is now the single canonical guide — coverage map, host/driver roles, and worked examples end to end. - [2026-06-12: Agent-First HTTP v0.7.0: One Flag for the Hard Sites, Brave by Default](https://agentfirstkit.com/agent-first-http/blog/2026-06-12-http-v0.7.0-takeover-by-default-brave.md): v0.7.0 collapses the hard-site dance into a single flag: `afhttp fetch --takeover`. It auto-discovers the local host, gives each site its own isolated profile by default, and only asks a human to step in when a wall actually blocks it. `container install` is now takeover-ready out of the box on Brave — built-in ad/tracker blocking and the browser a human drives, no `--with` flags to remember. Secrets are redacted from agent-visible output by default, with an explicit `--reveal-token-secret` to opt in. The whole release is one thing: make afhttp simpler for an agent to use and harder to use wrong. - [2026-06-07: Agent-First HTTP v0.6.0: A 200 Isn't Proof](https://agentfirstkit.com/agent-first-http/blog/2026-06-07-http-v0.6.0-bot-walls-become-data.md): v0.6.0 is about not getting fooled. afhttp now recognizes a bot wall or security challenge on the cheap HTTP path — Cloudflare, Turnstile, generic access-denied — surfaces it as a `page_kind` plus warning, and auto-escalates to a real browser instead of handing the agent a 200 that isn't the page. The fetch trace gets honest too: per-stage timing, `current_stage`, `capture_reason`, `wait_mode`, so an agent can see where a fetch spent its time and why it stopped. Plus readiness tuning, a `takeover prepare` subcommand, a named display-provider abstraction, and a skill rewrite that tells the agent to reach for afhttp first. - [2026-06-03: Agent-First HTTP v0.5.0: One Command to a Browser Host, in Any Runtime](https://agentfirstkit.com/agent-first-http/blog/2026-06-03-one-command-container-hosts.md): v0.5.0 adds `afhttp container`: bring up the browser host in one command — Docker, Podman, or Apple Container — from a recipe embedded in the binary, no source tree required. - [2026-06-02: Agent-First HTTP v0.5.0: When the Page Needs a Browser](https://agentfirstkit.com/agent-first-http/blog/2026-06-02-http-v0.5.0-browser-backed-acquisition.md): v0.5.0 turns afhttp from an HTTP client into a full URL-acquisition tool. A single `afhttp fetch` covers the whole range — a plain HTTP request when that works, a real browser when it doesn't — and returns the page plus structured artifacts (rendered HTML, a DOM observation, a screenshot, network and console logs) an agent can branch on. It adds a browser-host / agent-driver split, a raw CDP escape hatch, deep network capture, an ops panel with optional real-display takeover for human login/captcha/2FA, and persistent profiles. The public contract converged in the process: flat `*_file` artifact paths, one profile per host, and no legacy aliases. - [2026-05-23: Agent-First HTTP v0.4.3: Release Hardening for Agent Tools](https://agentfirstkit.com/agent-first-http/blog/2026-05-23-release-hardening-for-agent-tools.md): The v0.4.3 release moved afhttp into Agent-First Kit, tightened structured-output discipline, and hardened cross-platform packaging. - [2026-03-26: Agent-First HTTP v0.4: A Narrower Runtime with Complete Help](https://agentfirstkit.com/agent-first-http/blog/2026-03-26-narrow-runtime-complete-help.md): The v0.4 line removed the MCP server surface, generated CLI docs from clap definitions, and made --help complete for agents. - [2026-03-11: Agent-First HTTP v0.3.4: Output Formats Preserved Responses](https://agentfirstkit.com/agent-first-http/blog/2026-03-11-output-formats-preserved-responses.md): The v0.3.4 update extended yaml/plain rendering to pipe mode while protecting server response bodies from formatter reinterpretation. - [2026-03-08: Agent-First HTTP v0.3.2: Requests Became Previewable](https://agentfirstkit.com/agent-first-http/blog/2026-03-08-requests-became-previewable.md): The v0.3.2 update added dry-run previews and actionable hints, so agents can inspect an HTTP request shape before it touches the network. - [2026-02-21: Agent-First HTTP v0.3: One Request, One JSON Line](https://agentfirstkit.com/agent-first-http/blog/2026-02-21-first-release.md): The early Agent-First HTTP release line: a structured HTTP client for agents that turns requests, streaming bodies, and transport failures into stable JSON events. ## Agent-First Mail — Blog - [2026-11-02: Agent-First Mail v0.13.0: Past the Mailbox Server](https://agentfirstkit.com/agent-first-mail/blog/2026-11-02-mail-v0.13.0-past-the-mailbox-server.md): afmail no longer needs an IMAP server. Mail can arrive through Cloudflare Email Routing and R2 and leave through Cloudflare's sending API, messages from a web form can come in as plain JSON files, and an ordinary process can wake an agent only when there is mail to read. Plus a person deciding what happened to a send nobody can vouch for, and a 4,000-message mailbox that went from five and a half minutes per pull to about twenty seconds. - [2026-10-10: Agent-First Mail v0.12.0: Where the Link Goes](https://agentfirstkit.com/agent-first-mail/blog/2026-10-10-mail-v0.12.0-where-the-link-goes.md): HTML mail used to reach the Review page as walls of tracking URLs or as text with every link target stripped. It is read through a Markdown copy now, and every link says which domain it goes to and whether that is the sender's. Plus a cleared Cc that was still sent, a move that waited forever for a search index, and a rejection that is no longer mistaken for an unknown. - [2026-08-24: Agent-First Mail v0.11.0: Nothing Waits for Done](https://agentfirstkit.com/agent-first-mail/blog/2026-08-24-mail-v0.11.0-nothing-waits-for-done.md): Review used to stage a person's answers and commit them all at once when they pressed Done, which meant closing the window threw the work away. Each item now settles the moment it is decided, the workspace is the source of truth rather than the process, and a draft can no longer attach a file it merely points at. - [2026-08-07: Agent-First Mail v0.10: Not Our Window](https://agentfirstkit.com/agent-first-mail/blog/2026-08-07-mail-v0.10.0-not-our-window.md): v0.10 hands the human review window to agent-first-ui: afmail deletes its own listener, request parser, and browser launcher, and gains what it never had — four independently replaceable review surfaces, a frontend contract with a version on it, and doctor checking the UI a person will actually see. Credentials in config move to one key with the AFDATA source grammar. - [2026-08-05: Agent-First Mail v0.9: Every Legal Combination](https://agentfirstkit.com/agent-first-mail/blog/2026-08-05-mail-v0.9.0-every-legal-combination.md): v0.9 compiles the whole afmail CLI from a closed registry — 106 commands, 140 legal argument shapes, one --help call to see all of a command's at once — and hardens what the agent acts on underneath it: a message's identity is now its bytes, a push refuses stale UIDs, and an edit to a file you wrote moves only the keys that changed. - [2026-07-30: Agent-First Mail v0.8: Ask the Human](https://agentfirstkit.com/agent-first-mail/blog/2026-07-30-mail-v0.8.0-ask-the-human.md): v0.8 brings the human decision inside afmail: interactive review windows for triage, cases, notifications, and drafts, opened by the agent and answered by you — rendered from the same typed documents the agent reads, through one MiniJinja template system you can edit and translate. - [2026-07-14: Agent-First Mail v0.7: Run It Again](https://agentfirstkit.com/agent-first-mail/blog/2026-07-14-mail-v0.7.0-run-it-again.md): v0.7 makes every mutation safe to retry: a --request-id turns any local change into an idempotent operation, remote effects only fire from a plan you already reviewed, and pulls run inside a byte-and-count budget you set up front. - [2026-07-03: Agent-First Mail v0.6: Fewer Doors](https://agentfirstkit.com/agent-first-mail/blog/2026-07-03-mail-v0.6.0-fewer-doors.md): v0.6 collapses the separate archive subcommand into case, replaces archive/reopen cycles for contacts and cases with explicit delete, and splits message restore into unspam and untrash — fewer places to look, more precise actions. - [2026-07-01: Agent-First Mail v0.5: Notifications Aren't Cases](https://agentfirstkit.com/agent-first-mail/blog/2026-07-01-mail-v0.5.0-notifications.md): v0.5 splits the two things afmail used to call 'archive': finished cases stay in the archive, but standalone notification and reference mail now files into its own notification collections — and every case has to name the queue it belongs to. - [2026-06-28: Agent-First Mail v0.4: The Workspace Speaks Your Language](https://agentfirstkit.com/agent-first-mail/blog/2026-06-28-mail-v0.4.0-language.md): v0.4 localizes the part of afmail you actually read: every generated Markdown view and workspace scaffold now ships in nine fully-translated language packs, chosen by one config line — while the CLI, the wire format, and the one push boundary stay exactly where they were. - [2026-06-20: Agent-First Mail v0.3: The Workspace Knows Who](https://agentfirstkit.com/agent-first-mail/blog/2026-06-20-mail-v0.3.0-who.md): v0.3 gives the mailbox workspace a sense of who: contact cards that auto-link to cases and materialise onto message views, and an identity registry of named send personas your drafts can write footers from — both still behind the one push boundary. - [2026-06-17: Agent-First Mail v0.2: One Push, All or Nothing](https://agentfirstkit.com/agent-first-mail/blog/2026-06-17-mail-v0.2.0-one-push.md): v0.2 tightens the push boundary into a single all-or-nothing operation, adds a draft preview, unifies message restore and lifecycle-agnostic case reads, lets you archive a message directly, and lets a workspace live in a subdirectory. - [2026-06-06: Agent-First Mail v0.1: Your Inbox as a Git Worktree](https://agentfirstkit.com/agent-first-mail/blog/2026-06-06-first-release.md): Agent-First Mail gives each mailbox a local, file-first workspace: pull mail into files, work cases and drafts offline, and push remote effects only when you explicitly confirm. ## Agent-First Pay — Blog - [2026-10-09: Agent-First Pay v0.11.0: Not Knowing Is Not Failing](https://agentfirstkit.com/agent-first-pay/blog/2026-10-09-pay-v0.11.0-not-knowing-is-not-failing.md): Every send that errored was treated as a send that never happened: the plan released, the idempotency key cleared, the next retry free to pay again. A lost reply is not a failed payment. afpay now has a third outcome — unknown — and keeps the key held until someone checks. Plus endpoint failover that could pay twice, fees nobody quoted, a 200 that was not a receipt, and backup passwords in the process table. - [2026-08-24: Agent-First Pay v0.10.0: An Outage Is Not Consent](https://agentfirstkit.com/agent-first-pay/blog/2026-08-24-pay-v0.10.0-an-outage-is-not-consent.md): When every provider's quote failed, the send confirmation printed 'skipping confirmation' and returned true — the payment went out having asked nobody. That, a rounding bias that made split payments free against the global cap, and an allowlist that let one configured category start a wide-open public daemon. Plus gRPC and its hand-rolled crypto layer are gone, replaced by a resource API with an OpenAPI face. - [2026-08-03: Agent-First Pay v0.9.0: Illegal Commands Stopped Existing](https://agentfirstkit.com/agent-first-pay/blog/2026-08-03-pay-v0.9.0-illegal-commands-stopped-existing.md): v0.9.0 compiles the whole afpay CLI from a closed-world registry: one root, 27 command groups, 89 leaves, generated from a table rather than written out five times. Rules that used to be runtime checks — 'this credential is required for that backend', 'this listener flag belongs to that mode' — are now shapes, so an agent reads the requirement out of the help instead of discovering it by failing. Discovery answers cli-help-v2 JSON, scoped per command and bounded in size; bare --version answers a version event. Arguments become command-local, several aliases are gone, and the container image finally catches up to phoenixd 0.9.0 and Bitcoin Core 31.1 — including a signet path that never actually started. - [2026-07-22: Agent-First Pay v0.8.0: Retry Without Paying Twice](https://agentfirstkit.com/agent-first-pay/blog/2026-07-22-pay-v0.8.0-retry-without-paying-twice.md): v0.8.0 rebuilds afpay around the one question an autonomous payer cannot answer for itself: 'did that send go through?' An agent-supplied idempotency key makes a resend replay the first result instead of paying again; spend reservations hold the money in flight and expose reservation_ids for later reconciliation; EVM chain_id and Solana cluster are pinned at send time and fail closed on mismatch; and structured errors carry retry_after_ms so an agent can pace its own backoff. Underneath, the runtime is reorganized into six modes — cli, pipe, interactive, tui, rpc, rest — that all answer the same code-tagged protocol, and every one can be asked for its own schema. - [2026-05-23: Agent-First Pay v0.7: Payment Tools Joined Agent-First Kit](https://agentfirstkit.com/agent-first-pay/blog/2026-05-23-agentfirstkit-payment-hardening.md): The v0.7 release moved afpay into Agent-First Kit, hardened security and remote flows, kept backup/TUI work, and refreshed major wallet dependencies. - [2026-05-23: Agent-First Pay: A Payment Tool Designed from the Agent Side](https://agentfirstkit.com/agent-first-pay/blog/2026-05-23-agent-payment-tool.md): What a multi-chain payment tool should look like when the primary caller is an agent: human-set spend limits, stable settlement ids, one interface across chains, structured receive payloads, and durable redacted audit. - [2026-03-25: Agent-First Pay v0.6: Wallet Operations Got a TUI and Backups](https://agentfirstkit.com/agent-first-pay/blog/2026-03-25-wallet-ops-got-ui-and-backups.md): The v0.6 release added a full-screen wallet TUI, backup/restore commands, copyable payment artifacts, and a modular runtime shared across modes. - [2026-03-18: Agent-First Pay v0.4.2: Solana Payments Became Order-Bound](https://agentfirstkit.com/agent-first-pay/blog/2026-03-18-solana-reference-keys.md): The v0.4.2 update added Solana reference key support so agents can bind send and receive flows to a specific order or intent. - [2026-03-16: Agent-First Pay v0.4: Security Became Part of the Runtime](https://agentfirstkit.com/agent-first-pay/blog/2026-03-16-security-became-runtime.md): The v0.4 release tightened REST security, standard wallet compatibility, rate limits, FX freshness, provider fallback, and wallet storage. - [2026-03-14: Agent-First Pay v0.2: Settlement IDs Became Real](https://agentfirstkit.com/agent-first-pay/blog/2026-03-14-real-settlement-ids.md): The v0.2 release made receive --wait flows return real on-chain transaction IDs and improved reproducible self-hosted payment deployments. - [2026-03-11: Agent-First Pay v0.1: Multi-Chain Payments for AI Agents](https://agentfirstkit.com/agent-first-pay/blog/2026-03-11-first-release.md): The first Agent-First Pay release line: a single agent-oriented payment interface that spans Cashu, Lightning, Solana, EVM, and Bitcoin with spend limits and JSONL output. ## Agent-First PDF — Blog - [2026-10-20: Agent-First PDF: The Article, Not the Page](https://agentfirstkit.com/agent-first-pdf/blog/2026-10-20-the-article-not-the-page.md): Text pulled out of a PDF comes back page-shaped: lines wrapped where the layout wrapped them, a page number every forty lines, columns read across instead of down. Agent-First PDF recovers the article underneath, keeps the evidence it was derived from, and reports every page it could not read. - [2026-10-20: Agent-First PDF v0.1.0](https://agentfirstkit.com/agent-first-pdf/blog/2026-10-20-first-release.md): The first release of Agent-First PDF — article text, structure and images recovered from PDF layouts, with the engine's evidence kept beside every interpretation, one child process per document by default, and a pinned PDFium build installed and verified on request. ## Agent-First PSQL — Blog - [2026-11-01: Agent-First PSQL v0.13.0: Read-Only Is a Transaction Mode](https://agentfirstkit.com/agent-first-psql/blog/2026-11-01-psql-v0.13.0-read-only-is-a-transaction-mode.md): afpsql-readonly let `;COMMIT` through, accepted ProxyJump hops that could reopen command execution, and checked a pipe session's container runtime without the environment it would connect with. All three are closed, and the docs now say plainly what a read-only transaction does not stop. Plus explicit connection fields that beat environment DSNs, secrets kept out of parse errors, and a reconciled storage inspection report. - [2026-10-10: Agent-First PSQL v0.12.0: The Host You Named](https://agentfirstkit.com/agent-first-psql/blog/2026-10-10-psql-v0.12.0-the-host-you-named.md): A refused SSH setting used to fall through to a direct connection, reaching whatever answered on the local port instead of the database behind SSH. A lost transaction used to drop its queued statements into autocommit. Both now stop where they are. Plus a streamed one-shot query that could deadlock, a pipe that could stop reading stdin, and SSH options cut to an allowlist. - [2026-08-24: Agent-First PSQL v0.11.0: A Cancel Is a Request](https://agentfirstkit.com/agent-first-psql/blog/2026-08-24-psql-v0.11.0-a-cancel-is-a-request.md): Cancelling a query took its terminal event slot before the CancelRequest was even sent, so a write that had already committed came back as 'cancelled' — the worst answer this program can give. That, a COMMIT sent down a brand-new connection after the backend died, and inspection panels rebuilt around objects instead of the snapshot query's row shape. - [2026-08-07: Agent-First PSQL v0.10.0: The Window and the Agent Cannot Disagree](https://agentfirstkit.com/agent-first-psql/blog/2026-08-07-psql-v0.10.0-the-window-and-the-agent-cannot-disagree.md): Five panels that open a window onto PostgreSQL — four you read, one you approve. Each runs the exact SQL its inspect sibling runs, an approved statement runs the ordinary path, and only an approval runs anything: a closed window, a refusal and a lapsed credential are the same answer. - [2026-08-05: Agent-First PSQL v0.9.1: The Errors That Are Not Yours](https://agentfirstkit.com/agent-first-psql/blog/2026-08-05-psql-v0.9.1-the-errors-that-are-not-yours.md): v0.9.0 made every illegal invocation an exit-2 rejection that says «rewrite your command line». v0.9.1 is about the two failures where the caller has nothing to rewrite: a handler that reads an argument its own shape never declared, which used to become an empty string that looked like a value someone passed, and an output sink that could not be opened, which used to be reported as a usage error. Both are exit 1 with a code of their own, and a new test drives all 23 shapes through their handlers so the first one is caught before it ships. Rejections also stopped quoting the token back, so a psql-style `-d` typo can no longer echo its password into a logged error event. Picks up Agent-First Data 0.31.0. - [2026-08-03: Agent-First PSQL v0.9.0: Making the Illegal Invocation Unwritable](https://agentfirstkit.com/agent-first-psql/blog/2026-08-03-psql-v0.9.0-illegal-invocations-become-unwritable.md): v0.9.0 compiles the CLI from a closed registry of 17 commands and 23 shapes, so an invocation runs only when it matches one of them and every combination that used to be caught by a runtime check — or quietly ignored — is now an unknown argument at parse time. The nine `--*-secret[-env|-config]` flags collapse into three typed sources (`--dsn`, `--conninfo`, `--password`) that accept a literal, `env:NAME`, or `file:PATH#DOT_PATH`. Container flags name their driver, so an option a driver does not have cannot be written. And the write boundary the README always promised is now actually enforced inside pipe transactions, where three JSONL lines with no permission field could previously delete rows. - [2026-07-29: Agent-First PSQL v0.8.2: A Test That Poisoned Its Neighbours](https://agentfirstkit.com/agent-first-psql/blog/2026-07-29-psql-v0.8.2-a-test-that-poisoned-its-neighbours.md): Two tests mutated PGHOST and AFPSQL_DSN_SECRET inside the shared library test binary while unlocked readers ran in parallel, so five unrelated connection tests failed depending on thread scheduling. Both moved into a dedicated integration binary that owns process-environment mutation outright. Also picks up Agent-First Data 0.26.2. - [2026-07-29: Agent-First PSQL v0.8.1: The Release Check Reads the Right Stream](https://agentfirstkit.com/agent-first-psql/blog/2026-07-29-psql-v0.8.1-release-smoke-reads-the-right-stream.md): v0.8.0's own release smoke test still read errors from stdout, so it failed the binary build after publishing and shipped v0.8.0 without downloadable binaries. v0.8.1 fixes the check — including a secret-leak assertion that had started passing against an empty string — and moves it into the release gate so it runs before anything irreversible. - [2026-07-29: Agent-First PSQL v0.8.0: The DSN Crosses the Boundary Intact](https://agentfirstkit.com/agent-first-psql/blog/2026-07-29-psql-v0.8.0-dsn-across-boundaries-and-output-routing.md): v0.8.0 lets `--dsn-secret` and `--conninfo-secret` work with `--ssh` and `--container`: afpsql parses the connection string locally, derives only the transport's internal endpoint from it, and carries authentication, database, startup options, timeouts, keepalives, channel binding, and TLS across the tunnel unchanged — with the SSH stdio bridge now encrypted too. Event routing now follows the invocation: a finite query splits result to stdout and diagnostics to stderr, while `--mode pipe` and `--stream-rows` keep their ordered stream whole on stdout. The canonical CLI surface goes long-flags-only so psql's shorts keep their psql meanings. - [2026-07-19: Agent-First PSQL v0.7.0: Read the Secret From the Config](https://agentfirstkit.com/agent-first-psql/blog/2026-07-19-psql-v0.7.0-config-secret-sources-and-readonly.md): v0.7.0 lets afpsql read a connection secret straight from the application's own JSON, TOML, YAML, or dotenv config file — `--dsn-secret-config FILE DOT_PATH` and its conninfo/password siblings — resolved once, in-process, with no `jq`/`yq` subprocess and no secret in argv or shell history. Configured secrets always render as `***` in runtime config output regardless of source, and `afpsql-readonly` is redefined as a PostgreSQL write guard rather than a host sandbox: it restores SQL files, config sources, SSH options, container runtimes, redirects, and skill management while still refusing writes. - [2026-06-01: Agent-First PSQL v0.6.3: Look Before You Touch](https://agentfirstkit.com/agent-first-psql/blog/2026-06-01-psql-v0.6.3-inspect-dry-run-explain-and-transactions.md): v0.6.3 gives an agent everything it needs to understand a database before changing it: `afpsql inspect` for schema discovery, `--dry-run` to prepare and validate a statement without running it, `--explain` / `--explain-analyze` for the query plan, and pipe-mode `begin`/`commit`/`rollback` for explicit multi-statement transactions with savepoint-isolated failures. It also soft-truncates oversized inline results instead of erroring, and ships correctness fixes for query cancellation, value decoding (bytea and arrays), and NUMERIC bind precision. - [2026-05-28: Agent-First PSQL v0.6.2: Container Transport Family and Self-Describing Sessions](https://agentfirstkit.com/agent-first-psql/blog/2026-05-28-psql-v0.6.2-container-transport-and-session-info.md): v0.6.2 generalizes the docker transport into a container transport family (podman, nerdctl, compose, kubectl) with structured scope flags and SSH chaining, adds a session_info pipe request so agents can introspect their session's transport, permission default, and limits instead of probing with failing queries, and surfaces two new log events for implicit behaviors that previously had to be inferred. - [2026-05-26: Agent-First PSQL v0.6.1: Embedded Skill Installer and SQLSTATE on Connect](https://agentfirstkit.com/agent-first-psql/blog/2026-05-26-psql-v0.6.1-skill-installer-and-connect-sqlstate.md): v0.6.1 ships the Agent-First PSQL skill inside the binary so Claude Code and Codex can install it with one command, and preserves PostgreSQL SQLSTATE plus message, detail, and hint on connect_failed so agents can distinguish auth, role, database, capacity, and startup failures without parsing prose. - [2026-05-25: Agent-First PSQL v0.6: SSH Transport and Explicit Write Permissions](https://agentfirstkit.com/agent-first-psql/blog/2026-05-25-psql-v0.6-ssh-and-permissions.md): v0.6 adds an SSH transport that keeps the agent local while reaching server-only PostgreSQL, and splits write permission into separate direct and SSH families so an agent cannot silently turn a read across a boundary into a remote write. - [2026-05-23: Agent-First PSQL v0.5: A PostgreSQL Connector Designed from the Agent Side](https://agentfirstkit.com/agent-first-psql/blog/2026-05-23-agent-postgres-connector.md): The v0.5 afpsql line asks what a PostgreSQL connector should look like when the primary caller is an agent: structured state, explicit permissions, stable sessions, and local control over remote databases. - [2026-03-26: Agent-First PSQL v0.4: A Native Runtime with Complete Help](https://agentfirstkit.com/agent-first-psql/blog/2026-03-26-native-runtime-complete-help.md): The v0.4 line removed MCP server mode, generated CLI docs from the source command definition, and made --help complete for agents. - [2026-03-11: Agent-First PSQL v0.3.1: Output Policy Protected SQL Rows](https://agentfirstkit.com/agent-first-psql/blog/2026-03-11-output-policy-protected-rows.md): The v0.3.1 update separated SQL payloads from output-layer redaction and preserved row structure across JSON, YAML, plain, and MCP responses. - [2026-03-08: Agent-First PSQL v0.2.1: SQL Became Previewable](https://agentfirstkit.com/agent-first-psql/blog/2026-03-08-sql-became-previewable.md): The v0.2.1 update added dry-run SQL previews, actionable error hints, and better config invalidation for stateful sessions. - [2026-02-22: Agent-First PSQL v0.1: SQL Queries as Structured Events](https://agentfirstkit.com/agent-first-psql/blog/2026-02-22-first-release.md): The first Agent-First PSQL release line: PostgreSQL rows, timing, and SQLSTATE failures as machine-readable events instead of terminal prose. ## Agent-First Slug — Blog - [2026-11-01: Agent-First Slug v0.9: Checked Before It Is Needed](https://agentfirstkit.com/agent-first-slug/blog/2026-11-01-slug-v0.9.0-checked-before-it-is-needed.md): A fallback that broke its own configuration used to fail only on the rare input that needed it. Fallbacks are now checked before any input is read, every afslug error names the rule it broke, decimal dots need digits the character set actually keeps, and errors identify invisible characters by code point. - [2026-10-08: Agent-First Slug v0.8: The Word It Cut in Half](https://agentfirstkit.com/agent-first-slug/blog/2026-10-08-slug-v0.8.0-the-word-it-cut-in-half.md): No character set could keep a Thai, Devanagari or Tamil word whole, and truncation could cut a character away from the marks that complete it. A new character set keeps combining marks where they belong, truncation drops a character whole instead of breaking it, and three configurations that used to be accepted are now refused. - [2026-08-23: Agent-First Slug v0.7: A Config That Describes Its Output](https://agentfirstkit.com/agent-first-slug/blog/2026-08-23-slug-v0.7.0-a-config-that-describes-its-output.md): Three rules let a SlugConfig describe something other than what came out of it: case mapping ran after filtering, a delimiter could be a character the filter kept, and a fallback was validated against the target surface but never against the configuration it stood in for. All three are fixed, and two of them change generated slugs. - [2026-08-02: Agent-First Slug v0.6: A Shape, Not a Sentence](https://agentfirstkit.com/agent-first-slug/blog/2026-08-02-slug-v0.6.0-a-shape-not-a-sentence.md): afslug's CLI is now compiled from a closed-world registry: one source for argv parsing, typed values, legal combinations, help, and docs/cli.md. clap is gone. The visible payoff is that a constraint which used to live in a help sentence and a runtime check is now a shape the parser enforces and `--help` lists. - [2026-07-28: Agent-First Slug v0.5: Discovery Without Guessing](https://agentfirstkit.com/agent-first-slug/blog/2026-07-28-slug-v0.5.0-discovery-without-guessing.md): An agent meeting a CLI for the first time does the same three things: asks what it can do, asks what version it is, and runs something. afslug answered the first two in prose an agent had to parse and the third by mixing results and errors into one stream. v0.5 makes all three structured. `afslug --help` returns a scoped result — this command level only, globals marked global, `--recursive` for the whole tree. `afslug --version` returns a result, not a sentence. And `--output-to` decides whether errors land on stderr beside the results or join them in one ordered stream. - [2026-07-21: Agent-First Slug v0.4: The Tool Installs Its Own Skill](https://agentfirstkit.com/agent-first-slug/blog/2026-07-21-slug-v0.4.0-installs-its-skill.md): A slug library is a few hundred lines; the hard part was never the transliteration table. v0.1 made every policy explicit at the call site so the code states which surface it targets. But explicit policy only pays off if the agent reaches for afslug and knows how it reasons — knowledge that used to sit in a README the agent might never open. v0.4 ships that knowledge as a skill the binary installs itself: `afslug skill install` drops the Agent-First Slug skill into Codex, Claude Code, opencode, or Hermes, bundling SKILL.md and its Codex agent interface as one tree. And `afslug --version` now answers in the shared structured protocol — name, display name, version, and the git SHA it was built from — the same shape every agent-first CLI reports. - [2026-06-25: Agent-First Slug v0.1.0](https://agentfirstkit.com/agent-first-slug/blog/2026-06-25-first-release.md): The first release of Agent-First Slug — a Rust slug library where every policy is set explicitly by the caller, with no hidden defaults that differ by surface. ## Agent-First Terminal — Blog - [2026-10-10: Agent-First Terminal v0.3.0](https://agentfirstkit.com/agent-first-terminal/blog/2026-10-10-a-terminal-that-stops-reading.md): A program that stops reading its input could freeze every other session, a client that stops reading events could freeze the server, and a supervisor's SIGTERM was not a stop. Writes, subscriptions and shutdown are now bounded by their own owner, an SSE resume says when history is missing, and a secret window no longer leaves its screen behind. - [2026-08-27: Agent-First Terminal v0.2.0](https://agentfirstkit.com/agent-first-terminal/blog/2026-08-27-answering-the-terminal.md): The Windows build shipped as a window that never printed anything, because a terminal is a conversation and this one only listened — ConPTY opens by asking where the cursor is and says nothing until it is told. - [2026-08-26: Agent-First Terminal v0.1.0](https://agentfirstkit.com/agent-first-terminal/blog/2026-08-26-first-release.md): The first release of Agent-First Terminal — a multi-session PTY runtime an agent can drive and a person can watch, with input leases that a human preempts, a secret mode that publishes nothing derived from what was typed, and the same runtime reachable as a window, a LAN link, or a registered session. ## Agent-First UI — Blog - [2026-11-01: Agent-First UI v0.9.0: Trust Belongs to One Checkout](https://agentfirstkit.com/agent-first-ui/blog/2026-11-01-ui-v0.9.0-trust-belongs-to-one-checkout.md): Trusting a workspace frontend now trusts that directory and no other, so two checkouts no longer revoke each other. A remote Provider can take as long to answer as the viewer is willing to wait, a minimised macOS window no longer ends its question, and the remote view and ask page speak Traditional Chinese. - [2026-10-31: Agent-First UI v0.8.0: A Credential Stays Where It Was Handed](https://agentfirstkit.com/agent-first-ui/blog/2026-10-31-ui-v0.8.0-a-credential-stays-where-it-was-handed.md): An AFUI session URL is its own credential, and this release follows it everywhere it went: process arguments, Referer headers, redirects, debug output, and streams that outlived a revocation. A question can now show images, tables, and diffs as evidence, and every ending of a session goes through one slot. - [2026-10-24: Agent-First UI v0.7.0: A Question Without a Page](https://agentfirstkit.com/agent-first-ui/blog/2026-10-24-ui-v0.7.0-a-question-without-a-page.md): `afui ask` puts a few questions with suggested answers in front of a person and returns one structured result, with no Provider, page, or button bindings to write. Notes sit on the choice they qualify, a written answer is one more card, and a caller whose commands time out can start a question, walk away, and collect the answer later. - [2026-10-09: Agent-First UI v0.6.0: Remote Until Proven Local](https://agentfirstkit.com/agent-first-ui/blog/2026-10-09-ui-v0.6.0-remote-until-proven-local.md): A page can now open a link outside itself, in the system browser for a local window and in a new tab for a remote view. Deciding which case it was is the whole risk, so a surface counts as remote unless it presents a key only AFUI's own window holds. Also: sessions that close the way they say, streams that end when their attention budget does, and a registry that is written atomically on every platform. - [2026-08-23: Agent-First UI v0.5.0: The Boundary Moves to Where the Copies Were](https://agentfirstkit.com/agent-first-ui/blog/2026-08-23-ui-v0.5.0-the-boundary-moves-to-where-the-copies-were.md): v0.4.0 shipped a channel that refused to mean anything, and every page then built the same meaning on top of it. Roughly seven hundred lines of page runtime, four private event streams, four hand-written content policies and five wildcard error arms later, the measurement is unambiguous: AFUI now owns the typed session runtime, the page kernel, page composition, delivery facts and credential routing — plus three places where its own contract said something the code did not do. - [2026-08-13: Agent-First UI v0.4.0: A Session Is a Lifetime, Not a Question](https://agentfirstkit.com/agent-first-ui/blog/2026-08-13-ui-v0.4.0-a-session-is-a-lifetime-not-a-question.md): One interface, one person, one answer was almost right. The part that was wrong is that some interfaces are not a question: they show changing state while a person acts on it. v0.4.0 adds a bounded full-duplex channel that carries Provider values and refuses to mean anything by them, and finishes the delivery boundary around it — one page assembly, one delivery call, and an ending every delivery can express. - [2026-08-07: Agent-First UI v0.3.0: Everything Except the Interface](https://agentfirstkit.com/agent-first-ui/blog/2026-08-07-ui-v0.3.0-everything-except-the-interface.md): Handing a person a UI is not a rendering problem, it is a delivery problem: a credential, a lifetime, an origin, a window, a cleanup, and an answer that comes back typed. v0.3.0 replaces the view-hint protocol with a library that owns exactly that boundary and never sees a domain value. - [2026-07-14: Agent-First UI v0.2.0: One Space, Many Views](https://agentfirstkit.com/agent-first-ui/blog/2026-07-14-ui-v0.2.0-one-space-many-views.md): v0.2.0 reworks the protocol from a single surface a tool draws into a shared live space that each participant sees as their own projection. Owners are self-generated ids, capabilities and facts project per-participant, whispers reach one viewer, and a separate advisory view-hint layer suggests layout without ever running code. - [2026-06-16: Agent-First UI v0.1.0: Describe the Interface, Don't Ship It](https://agentfirstkit.com/agent-first-ui/blog/2026-06-16-ui-v0.1.0-describe-dont-ship.md): The first release of Agent-First UI — a protocol where a tool describes its interface as data instead of shipping UI code. A trusted host you own renders the screens and actions, redacts secrets, and gates risky operations, so nothing the tool sends can run on your machine. v0.1.0 ships the wire protocol, a JSON schema, conformance fixtures, and a lightweight Rust SDK.