{"content":[{"text":"total 16\ndrwxr-xr-x 3 root root 4096 Sep 18 08:00 .\ndrwxr-xr-x 7 root root 4096 Sep 18 21:23 ..\ndrwxr-xr-x 8 root root 4096 Sep 18 08:00 .git\n-rw-r--r-- 1 root root  408 Sep 18 08:00 .gitignore\n---\n","type":"text"}]}

{"content":[{"text":"200\n---\ntitle: The agent-native standard\ndescription: 'One bar for agent-readiness, measured on two surfaces: your CLI and your website. Scored MUST / SHOULD / MAY, not asserted.'\nurl: https://anc.dev/\n---\n\n# The agent-native standard\n\nOne bar for agent-readiness, measured on two surfaces: your CLI and your website. Scored MUST / SHOULD / MAY, not asserted.\n\nTo audit a CLI, install the linter (`anc`) locally: see [/install](https://anc.dev/install). To audit a website and its MCP server, use\n[/audit](https://anc.dev/audit) or the `audit_website` MCP tool. To install the agent-native-cli **skill bundle** (the Claude Code /\nCodex / Cursor / OpenCode skill), see [/skill](https://anc.dev/skill). Agents that speak MCP can connect directly to the\nstreamable-HTTP server at [https://anc.dev/mcp](https","type":"text"}]}

{"content":[{"text":"=== install ===\n---\ntitle: Install anc\ndescription: '`anc` is the reference linter for the agent-native CLI standard. It scores any CLI against the eight principles and tells you, by audit ID, where it passes and where it falls…'\nurl: https://anc.dev/install\n---\n\n# Install anc\n\n`anc` is the reference linter for the agent-native CLI standard. It scores any CLI against the eight principles and\ntells you, by audit ID, where it passes and where it falls short. Install it locally, then point it at a binary or a\nproject directory.\n\n## Homebrew\n\n```bash\nbrew install brettdavies/tap/agentnative\n```\n\n## Cargo\n\n```bash\ncargo install agentnative\n```\n\nFor a prebuilt binary without compiling from source (requires\n[`cargo-binstall`](https://github.com/cargo-bins/cargo-binstall); skip if you don't already have it):\n\n```bash\ncargo binstall agentnative\n```\n\n## GitHub Releases\n\nPlatform archives, including Windows builds and SHA256 checksums, live at\n[github.com/brettdavies/agentnative-cli/releases](https://github.com/brettdavies/agentnative-cli/releases). Download the\narchive for your platform, extract, and put the `anc` binary on `$PATH`.\n\n## What's next\n\nOnce installed, invoke the CLI as `anc`. See [/audit](https://anc.dev/audit) for usage: flags, output shapes, and how to interpret the\nper-principle audit IDs. The principles themselves are spelled out on [the homepage](https://anc.dev/), with one page per principle\n(`/p1` through `/p8`).\n\nTo install the **agent-native-cli skill bundle** instead (the Claude Code / Codex / Cursor / OpenCode skill that teaches\nan agent to write CLIs against this standard), see [/skill](https://anc.dev/skill).\n=== audit ===\n---\ntitle: Audit a CLI tool or a website\ndescription: 'anc.dev audits the two surfaces an agent works through: the command-line tools it runs and the websites it reads. Pick CLI or Website, enter a target, and the audit runs live…'\nurl: https://anc.dev/audit\n---\n\n# Audit a CLI tool or a website\n\nanc.dev audits the two surfaces an agent works through: the command-line tools it runs and the websites it reads. Pick\nCLI or Website, enter a target, and the audit runs live and lands on a shareable scorecard.\n\nThe audit runs live from this page in a browser. From an agent, use the MCP tools named below.\n\n## Audit a CLI tool\n\n`anc` is the reference linter for this standard. It scores any CLI tool against the eight principles and tells you, by\naudit ID, where it passes and where it falls short. The live audit above installs a tool in a sandbox and runs its\nbinary and behavioral audits; for source and project depth, run `anc` locally.\n\n## Install\n\nSee [/install](https://anc.dev/install) for `brew`, `cargo`, and platform-archive instructions. Once `anc` is on `$PATH`, the rest of\nthis page is what to do with it.\n\n## Run it\n\n```bash\n# Against the current project (cargo workspace, binary, or source tree)\nanc audit .\n\n# Against a compiled binary directly\nanc audit ./target/release/mycli\n\n# Agent-friendly output\nanc audit . --output json\n\n# Narrow to one principle\nanc audit . --principle 3\n```\n\n## Read the output\n\n```text\nP1 — Non-Interactive by Default\n  [PASS] Non-interactive by default (p1-non-interactive)\n  [PASS] Has --no-interactive flag (p1-flag-existence)\n\nP3 — Progressive Help\n  [PASS] Help flag produces useful output (p3-help)\n  [PASS] Version flag produces version output (p3-version)\n\nP4 — Fail Fast with Actionable Errors\n  [PASS] Rejects invalid arguments (p4-bad-args)\n\nP6 — Composable Command Structure\n  [PASS] Handles SIGPIPE cleanly (p6-sigpipe)\n  [PASS] Respects NO_COLOR (p6-no-color-behavioral)\n\nP7 — Bounded, High-Signal Responses\n  [PASS] Has --quiet flag (p7-quiet)\n=== skill ===\n# Install agent-native-cli\n\n> Build CLI tools that AI agents can operate reliably.\n\nOne skill, one repo. Choose your host below, run the command, the agent picks up `SKILL.md` on next launch. The same machine-readable manifest is at [`/skill.json`](https://anc.dev/skill.json).\n\n## Choose your host\n\n### Claude Code\n\n```bash\ngit clone --depth 1 https://github.com/brettdavies/agentnative-skill.git ~/.claude/skills/agent-native-cli\n```\n\n### Codex\n\n```bash\ngit clone --depth 1 https://github.com/brettdavies/agentnative-skill.git ~/.codex/skills/agent-native-cli\n```\n\n### Cursor\n\n```bash\ngit clone --depth 1 https://github.com/brettdavies/agentnative-skill.git ~/.cursor/skills/agent-native-cli\n```\n\n### Factory Droid\n\n```bash\ngit clone --depth 1 https://github.com/brettdavies/agentnative-skill.git ~/.factory/skills/agent-native-cli\n```\n\n### Kiro\n\n```bash\ngit clone --depth 1 https://github.com/brettdavies/agentnative-skill.git ~/.kiro/skills/agent-native-cli\n```\n\n### OpenCode\n\n```bash\ngit clone --depth 1 https://github.com/brettdavies/agentnative-skill.git ~/.config/opencode/skills/agent-native-cli\n```\n\n## What this does\n\nClones `https://github.com/brettdavies/agentnative-skill.git` into your host's skills directory. `.git/` is preserved so future updates are a `git pull`.\n\n## Already installed?\n\nIf the destination directory already holds this skill (origin matches), update in place:\n\n```bash\ncd <install-dir> && git pull --ff-only\n```\n\nIf it holds something else, remove it first, then re-run the install command:\n\n```bash\nrm -rf <install-dir>\n=== pricing ===\n# Not found\n\nThis path is not a page on this site. Recover from the machine index:\n\n- [Sitemap](https://anc.dev/sitemap.xml)\n- [llms.txt](https://anc.dev/llms.txt)\n=== llms.txt ===\n# The agent-native standard\n\n> One bar for agent-readiness, measured on two surfaces: your CLI and your website. Scored MUST / SHOULD / MAY, not asserted.\n\n## Programmatic access\n\n- [MCP server (streamable HTTP)](https://anc.dev/mcp)\n- [MCP server card (SEP-1649)](https://anc.dev/.well-known/mcp/server-card.json)\n- [MCP client skill](https://anc.dev/mcp-skill.md)\n\nNo authentication. Connect now: POST JSON-RPC to `https://anc.dev/mcp` — legacy clients send `initialize` with client protocolVersion `2025-06-18`; modern clients use `MCP-Protocol-Version: 2026-07-28`, SEP-2243 headers, and `_meta` in params (no `initialize`). Declared server revision: `2026-07-28`.\n\nFull recipes: [MCP client skill](https://anc.dev/mcp-skill.md).\n\n## Principles\n\n- [P1: Non-Interactive by Default](https://anc.dev/p1.md)\n- [P2: Structured, Parseable Output](https://anc.dev/p2.md)\n- [P3: Progressive Help Discovery](https://anc.dev/p3.md)\n- [P4: Fail Fast with Actionable Errors](https://anc.dev/p4.md)\n- [P5: Safe Retries and Explicit Mutation Boundaries](https://anc.dev/p5.md)\n- [P6: Composable and Predictable Command Structure](https://anc.dev/p6.md)\n- [P7: Bounded, High-Signal Responses](https://anc.dev/p7.md)\n- [P8: Discoverable Through Agent Skill Bundles](https://anc.dev/p8.md)\n\n## Pages\n\n- [Audit a CLI tool or a website](https://anc.dev/audit.md)\n- [Install anc](https://anc.dev/install.md)\n- [About this standard](https://anc.dev/about.md)\n- [Agent-native badge](https://anc.dev/badge.md)\n- [Changelog](https://anc.dev/changelog.md)\n- [Contribute](https://anc.dev/contribute.md)\n- [Methodology](https://anc.dev/methodology.md)\n- [Scorecard schema](https://anc.dev/scorecard-schema.md)\n- [Web scorecard schema](https://anc.dev/web-scorecard-schema.md)\n- [Using anc.dev's MCP server](https://anc.dev/mcp-skill.md)\n- [anc.dev MCP server](https://anc.dev/mcp.md)\n\n## Skill\n\n- [Skill (HTML)](https://anc.dev/skill.md)\n- [Skill (canonical JSON)](https://anc.dev/skill.json)\n\n## Scorecards\n\n- [Leaderboard](https://anc.dev/scorecards.md)\n- [Coverage Matrix](https://anc.dev/coverage.md)\n- [act](https://anc.dev/score/act/md)\n- [actionlint](https://anc.dev/score/actionlint/md)\n- [age](https://anc.dev/score/age/md)\n- [anc](https://anc.dev/score/anc/md)\n- [ast-grep](https://anc.dev/score/ast-grep/md)\n- [atuin](https://anc.dev/score/atuin/md)\n- [aws-cli](https://anc.dev/score/aws-cli/md)\n- [bandwhich](https://anc.dev/score/bandwhich/md)\n- [bat](https://anc.dev/score/bat/md)\n- [biome](https://anc.dev/score/biome/md)\n- [bird](https://anc.dev/score/bird/md)\n- [bottom](https://anc.dev/score/bottom/md)\n","type":"text"}]}

{"content":[{"text":"=== mcp-skill ===\n---\ntitle: Using anc.dev's MCP server\ndescription: anc.dev exposes the agent-native CLI standard catalog over a Model Context Protocol server at `https://anc.dev/mcp`. Thirteen tools cover five surfaces (registry, principles,…\nurl: https://anc.dev/mcp-skill\n---\n\n# Using anc.dev's MCP server\n\nanc.dev exposes the agent-native CLI standard catalog over a Model Context Protocol server at `https://anc.dev/mcp`.\nThirteen tools cover five surfaces (registry, principles, spec, scorecards, web audits) plus five resources for direct\nlookup. The catalog is public: no authentication, no API key.\n\n## Quick reference\n\nThe server speaks streamable HTTP per MCP spec revision `2026-07-28`. The endpoint serves **dual-stack** clients: legacy\n(`initialize` → `tools/call`) and modern (SEP-2243 headers + `_meta` in params, no `initialize`).\n\n### Legacy quick start\n\nDrive it from any MCP-aware client (Claude Code, Codex, Cursor) or raw JSON-RPC:\n\n```bash\n# 1. initialize the session (returns InitializeResult with instructions)\ncurl -sS https://anc.dev/mcp -H 'Content-Type: application/json' -d '{\n  \"jsonrpc\": \"2.0\", \"id\": 1, \"method\": \"initialize\",\n  \"params\": {\n    \"protocolVersion\": \"2025-06-18\",\n    \"capabilities\": {},\n    \"clientInfo\": {\"name\": \"demo\", \"version\": \"0.1\"}\n  }\n}'\n\n# 2. list every tool with its full input schema\ncurl -sS https://anc.dev/mcp -H 'Content-Type: application/json' -d '{\n  \"jsonrpc\": \"2.0\", \"id\": 2, \"method\": \"tools/list\"\n}'\n\n# 3. call a tool\ncurl -sS https://anc.dev/mcp -H 'Content-Type: application/json' -d '{\n  \"jsonrpc\": \"2.0\", \"id\": 3, \"method\": \"tools/call\",\n  \"params\": {\"name\": \"get_scorecard\", \"arguments\": {\"slug\": \"ripgrep\"}}\n}'\n```\n\n### Modern quick start (no initialize)\n\n```bash\nMETA='\"_meta\":{\"io.modelcontextprotocol/protocolVersion\":\"2026-07-28\",\"io.modelcontextprotocol/clientInfo\":{\"name\":\"demo\",\"version\":\"0\"},\"io.modelcontextprotocol/clientCapabilities\":{}}'\n\n# tools/list — Mcp-Name header omitted\ncurl -sS https://anc.dev/mcp \\\n  -H 'Content-Type: application/json' -H 'Accept: application/json' \\\n  -H 'MCP-Protocol-Version: 2026-07-28' -H 'Mcp-Method: tools/list' \\\n  -d \"{\\\"jsonrpc\\\":\\\"2.0\\\",\\\"id\\\":10,\\\"method\\\":\\\"tools/list\\\",\\\"params\\\":{$META}}\"\n\n# tools/call — Mcp-Name required\ncurl -sS https://anc.dev/mcp \\\n  -H 'Content-Type: application/json' -H 'Accept: application/json' \\\n  -H 'MCP-Protocol-Version: 2026-07-28' -H 'Mcp-Method: tools/call' -H 'Mcp-Name: get_scorecard' \\\n  -d \"{\\\"jsonrpc\\\":\\\"2.0\\\",\\\"id\\\":11,\\\"method\\\":\\\"tools/call\\\",\\\"params\\\":{\\\"name\\\":\\\"get_scorecard\\\",\\\"arguments\\\":{\\\"slug\\\":\\\"ripgrep\\\"},$META}}\"\n```\n\nAll examples below show the `arguments` object passed to `tools/call`; the JSON-RPC envelope around it is the same shape\nevery time.\n\n## Get a scorecard\n\nThe most-used surface. Two tools, three input shapes, one orchestration core. The split is honest about cost:\n`get_scorecard` is always cheap (registry-index or R2-cache lookup); `score_cli` may trigger a fresh container audit.\nBoth accept exactly one of `slug`, `binary`, `install`, or `github_url`.\n\n### \"I want the scorecard for a CLI I know is in the registry\"\n\nCall `get_scorecard` with the slug. A hit returns the result envelope: the same body `GET /score/<target>/json` serves,\nso the tool result and the page you follow it to cannot disagree.\n\n```json\n// tools/call get_scorecard { \"slug\": \"ripgrep\" }\n{\n  \"found\": true,\n  \"kind\": \"cli\",\n  \"tier\": \"registry\",\n  \"target\": \"ripgrep\",\n  \"scorecard_url\": \"https://anc.dev/score/ripgrep\",\n  \"markdown_url\": \"https://anc.dev/score/ripgrep/md\",\n  \"json_url\": \"https://anc.dev/score/ripgrep/json\",\n  \"freshness\": { \"cached\": true, \"scored_at\": null, \"refresh_after\": null },\n  \"spec_version\": \"2026.05\",\n  \"scorecard\": { \"...\": \"...\" },\n  \"score_pct\": 87,\n  \"anc_version\": \"0.7.2\"\n}\n```\n\nUse `tier` as your cost signal: `registry` means curated and committed, `cache` means a prior audit cached the result,\nand `live` means this call produced it. A curated result carries no `refresh_after`: it changes when the registry does,\nnot on a clock.\n\n### \"Is this binary in the live-score cache?\"\n\nSame tool, install command as input. A hit returns the cached scorecard with `tier: \"cache\"`; a target already being\naudited says so; a miss returns a typed redirect. None of the three is an error.\n\n```json\n// tools/call get_scorecard { \"install\": \"npm install -g cowsay\" }\n// HIT\n{\n  \"found\": true,\n  \"kind\": \"cli\",\n  \"tier\": \"cache\",\n  \"target\": \"cowsay\",\n  \"scorecard_url\": \"https://anc.dev/score/cowsay\",\n  \"markdown_url\": \"https://anc.dev/score/cowsay/md\",\n  \"json_url\": \"https://anc.dev/score/cowsay/json\",\n  \"freshness\": { \"cached\": true, \"scored_at\": \"2026-09-14T18:02:11.000Z\", \"refresh_after\": \"2026-09-21T18:02:11.000Z\" },\n  \"spec_version\": \"2026.05\",\n  \"scorecard\": { \"...\": \"...\" },\n  \"anc_version\": \"0.7.2\"\n}\n\n// ALREADY RUNNING\n{\n  \"found\": false,\n  \"in_progress\": true,\n  \"started_at\": \"2026-09-14T18:05:40.000Z\",\n  \"message\": \"an audit for this target is already running; poll this tool or read the result page shortly.\"\n}\n\n// MISS\n{\n  \"found\": false,\n  \"next_tool\": \"score_cli\",\n  \"spec_version\": \"2026.05\",\n  \"message\": \"no cached scorecard for this input. Call score_cli with the same arguments to run a fresh audit (subject to the audit rate limit and the operator-controlled live-scoring kill switch).\"\n}\n```\n\nThe miss is `isError: false`. Cache state is data, not failure. Follow the `next_tool` pointer. An `in_progress` answer\nmeans a run is already in flight for that target: wait for it rather than starting a second one.\n\n### \"I want to live-audit a CLI that isn't cached yet\"\n\nCall `score_cli` with the same input shape. On a registry or cache hit it redirects you back to `get_scorecard` (no\ncontainer run, no cost). On a true cache miss it runs a metered audit and returns the fresh scorecard.\n\n```json\n// tools/call score_cli { \"github_url\": \"https://github.com/owner/some-new-cli\" }\n// HIT — already cached, no audit ran\n{\n  \"audited\": false,\n  \"source\": \"live-cache\",\n  \"next_tool\": \"get_scorecard\",\n  \"kind\": \"cli\",\n  \"tier\": \"cache\",\n  \"target\": \"some-new-cli\",\n  \"scorecard_url\": \"https://anc.dev/score/some-new-cli\",\n  \"markdown_url\": \"https://anc.dev/score/some-new-cli/md\",\n  \"json_url\": \"https://anc.dev/score/some-new-cli/json\",\n  \"freshness\": { \"cached\": true, \"scored_at\": \"2026-09-14T18:02:11.000Z\", \"refresh_after\": \"2026-09-21T18:02:11.000Z\" },\n  \"spec_version\": \"2026.05\",\n  \"scorecard\": { \"...\": \"...\" },\n  \"message\": \"a cached live-score result already exists; call get_scorecard for the inline record.\"\n}\n\n// MISS — fresh container audit ran\n{\n  \"audited\": true,\n  \"source\": \"fresh-audit\",\n  \"kind\": \"cli\",\n  \"tier\": \"live\",\n  \"target\": \"some-new-cli\",\n  \"scorecard_url\": \"https://anc.dev/score/some-new-cli\",\n  \"markdown_url\": \"https://anc.dev/score/some-new-cli/md\",\n  \"json_url\": \"https://anc.dev/score/some-new-cli/json\",\n  \"freshness\": { \"cached\": false, \"scored_at\": \"2026-09-14T18:09:02.000Z\", \"refresh_after\": \"2026-09-21T18:09:02.000Z\" },\n  \"spec_version\": \"2026.05\",\n  \"scorecard\": { \"...\": \"...\" },\n  \"anc_version\": \"0.7.2\"\n}\n```\n\nA branch-scoped input (`owner/repo@branch`, or a `/tree/<branch>` URL) resolves to `/score/<owner>/<repo>@<branch>` and\nadds `source_sha`, the commit the sandbox cloned.\n\nThe tools are symmetric: `get_scorecard` returns `found: true` exactly when `score_cli` returns `audited: false` on the\nsame input. The cost difference (registry/cache lookup vs container run) is the only reason to choose between them.\n\n## Audit a website\n\nFour tools score a website and its MCP server against the same eight principles as a CLI, mirroring the scorecard\nsurface above. The web audit runs entirely as in-Worker network probes (HTTP, JSON-RPC over streamable-HTTP, CORS,\nDNS-over-HTTPS): no container, nothing crawled.\n\n- `get_website_audit` (cheap read): pass a `url`; returns `{ found: true, ...envelope }` with `kind: \"web\"` on a cache\n  hit, `{ found: false, in_progress: true, started_at }` while an audit for that host is already running, or\n  `{ found: false, next_tool: \"audit_website\" }` on a miss.\n- `audit_website` (metered fresh audit): runs a fresh audit and returns a single terminal envelope. There are no\n  progress notifications: the server runs stateless per-request. A cached result younger than one minute is returned\n  without re-running. Gated by `WEB_AUDIT_ENABLED` + `WEB_AUDIT_LIMITER_IP` (30 per hour per IP, no anon fallback).\n- `list_website_audits`: the website half of the leaderboard, curated by default; `view: \"all\"` adds the user-submitted\n  domains that opted in to public listing. Each entry carries `scorecard_url`.\n- `get_web_remediation`: the canonical fix for a web-audit `check_id`. Pass the failing row's `evidence` and it is\n  appended to the prompt as a delimited, length-bounded data block; omit it for the catalog text alone.\n\n```jsonc\n// tools/call audit_website { \"url\": \"anc.dev\" }\n```\n\n**Freshness.** Every result that carries a scorecard carries a `freshness` object beside it, outside the scorecard\nitself: `cached` is `true` for a served cache entry or a listing-only flag patch and `false` for a result the call\nproduced; `scored_at` is when the audit ran (`null` on a legacy entry with no stamp); `refresh_after` is `scored_at`\nplus the one-minute cache-reuse window. Past `refresh_after` a repeat call stops reusing the cached entry and tries a\nfresh audit, which is eligibility rather than a promise: the kill switch, the rate limits, and probe failures still\napply, and the cached scorecard is served as data when they refuse.\n\nWeb scorecards use the `score` / `results` / `coverage_summary` shape documented at\n[/web-scorecard-schema](https://anc.dev/web-scorecard-schema); each result page is at `anc.dev/score/<domain>`, named by the\nenvelope's own `scorecard_url`. The scorecard inside the envelope carries the read-time enrichment every surface\napplies: each row's current category and normative keyword, a `result` line, and an inline `remediation` object on\nevery non-passing row. `get_web_remediation` is for a check id you do not already hold a row for.\n\n\n### From a website result page\n\nA website result page publishes its own read-only [WebMCP](https://webmachinelearning.github.io/webmcp/) tools, so a\nbrowser agent reads a finished audit from the page it is already looking at. Each tool reads the rendered page and\nnothing else: no fetch, no form submission, no navigation. None of them starts an audit, so none of them can bypass the\nTurnstile challenge the browser audit sits behind. To run an audit, use the MCP tools above.\n\n| Tool                | Arguments                                        | Returns                                                                                                               |\n| ------------------- | ------------------------------------------------ | --------------------------------------------------------------------------------------------------------------------- |\n| `get_worksheet`     | `ids`, `keywords`, `statuses`, `offset`, `limit` | One row per matching finding: `id`, `keyword`, `tier`, `status`, `unprobed`, `result`, `remediable`.                  |\n| `get_fix_prompt`    | `id` (required)                                  | The stored prompt for one check id, or a reason it has none, or `found: false` for an id the page does not render.    |\n| `get_fix_prompts`   | `ids`, `keywords`, `statuses`, `offset`, `limit` | A prompt per matching fixable row; a selected row that needs no fix comes back with `remediable: false` and a reason. |\n| `get_audit_summary` | `offset`, `limit`                                | `site_score`, `global_score`, a count for each of the seven statuses, and the paged issue list.                       |\n\nEvery response is a JSON envelope carrying `ok`, the page's `cached` / `scored_at` / `refresh_after`, and the result.\nRejected input answers `{ \"ok\": false, \"error\": { \"code\", \"field\", \"message\" } }`, with `allowed` listing the accepted\nvalues when the field is an enum.\n\n**Filters.** `ids`, `keywords`, and `statuses` are independent dimensions. Values OR within one dimension and the\ndimensions AND across, so `{ \"keywords\": [\"must\", \"should\"], \"statuses\": [\"broken\"] }` selects the broken MUST and\nSHOULD rows. `keywords` accepts `must`, `should`, and `may`; `statuses` accepts `pass`, `noncompliant`, `broken`,\n`absent`, `n_a`, `skip`, and `error`. An omitted filter selects every value, with one exception: an omitted `statuses`\nselects the observed fixable rows (`broken`, `noncompliant`, and `absent`, excluding `unprobed` ones), because an agent\nasking for findings with no status in mind wants what it can fix. Naming statuses explicitly selects those rows whether\nor not the run probed them, so every rendered row stays reachable. A present but empty array is rejected rather than\nread as \"none\"; omit the filter instead.\n\n**Order.** Every paginated surface returns one order: normative keyword `must`, `should`, `may`; then status `broken`,\n`absent`, `noncompliant`, `error`, `pass`, `n_a`, `skip`; then observed rows before `unprobed` ones; then rendered page\norder.\n\n**Pagination.** `offset` defaults to 0 and accepts any integer from 0 up. `limit` defaults to 10 and accepts 1 through\n25. Each response reports `total` (items matching the filters), `returned` (items in this page), `omitted` (matching\nitems left after this page), and `next_offset`. Follow `next_offset` until it comes back `null`. A page carries whole\nitems only, so a page of long prompts can return fewer than `limit`; `next_offset` advances by `returned`, so following\nit never skips an item.\n\n**Prompt size.** Browser tools answer inside a 1,500-character cap. A prompt too long to fit whole comes back with\n`prompt_truncated: true` and a `full_fix` object naming both an MCP tool call (`get_web_remediation` with the check id)\nand the skill page's markdown URL. Only the `Fix:` line is ever shortened: the run's evidence block exists nowhere else\nin the response, and `Goal:` and `Skill:` are how a reader recovers everything else.\n\n## Browse the catalog\n\nThree tools over the curated registry. None of them require a network round trip on the server side, since every\nresponse is a slice of the build-time catalog projection. Live-scored binaries do **not** appear here; they only show up\nvia `get_scorecard` / `score_cli`.\n\n```json\n// tools/call list_tools\n[\n  {\n    \"slug\": \"ripgrep\", \"name\": \"ripgrep\", \"binary\": \"rg\",\n    \"install\": \"brew install ripgrep\",\n    \"version\": \"14.1.0\", \"score_pct\": 87,\n    \"scorecard_url\": \"/score/ripgrep\",\n    \"audit_profile\": null\n  },\n  \"...\"\n]\n\n// tools/call get_tool { \"slug\": \"ripgrep\" }\n{ \"found\": true, \"entry\": { \"...\": \"...\" } }\n\n// tools/call get_tool { \"slug\": \"nonexistent\" }\n{ \"found\": false, \"message\": \"no registry entry for slug: nonexistent\" }\n\n// tools/call search_tools { \"score_min\": 80, \"audit_profile\": \"default\" }\n[ \"...summaries matching all filters...\" ]\n```\n\nFilters AND together. Rows without a committed scorecard are excluded when either of `score_min` / `score_max` is set.\n`principle_min_score` is reserved for a future per-principle filter and is currently a no-op.\n\n## Read the spec\n\nTwo pairs of tools cover the spec text and the principles that derive from it. The principle records carry the\n`audit_id` strings the `anc` CLI emits. Those identifiers are useful when an agent is reading a scorecard and wants to\nlook up exactly which requirement a finding maps to.\n\n```json\n// tools/call list_principles\n[\n  {\n    \"n\": 1, \"slug\": \"p1-non-interactive-by-default\",\n    \"title\": \"Non-interactive by default\",\n    \"level_summary\": {\"must\": 3, \"should\": 2, \"may\": 1}\n  },\n  \"...\"\n]\n\n// tools/call get_principle { \"n\": 1 }\n{\n  \"found\": true,\n  \"principle\": {\n    \"n\": 1, \"slug\": \"p1-non-interactive-by-default\",\n    \"title\": \"Non-interactive by default\",\n    \"body_markdown\": \"...\",\n    \"requirements\": [\n      { \"id\": \"p1.r1\", \"level\": \"must\", \"summary\": \"...\", \"audit_ids\": [\"p1.r1.no-tty-prompt\"] },\n      \"...\"\n    ]\n  }\n}\n\n// tools/call list_spec_sections\n{\n  \"spec_version\": \"2026.05\",\n  \"sections\": [\n    { \"slug\": \"readme\", \"title\": \"README\", \"level\": 1, \"parent_slug\": null },\n    { \"slug\": \"p1-non-interactive-by-default\", \"title\": \"Non-interactive by default\", \"level\": 2, \"parent_slug\": \"principles\" },\n    \"...\"\n  ]\n}\n\n// tools/call get_spec_section { \"slug\": \"scoring\" }\n{\n  \"found\": true,\n  \"section\": {\n    \"slug\": \"scoring\", \"title\": \"Scoring\",\n    \"body_markdown\": \"...\",\n    \"spec_version\": \"2026.05\"\n  }\n}\n```\n\n`get_principle` and `get_spec_section` both return `isError: false` with `found: false` on a miss. Absence is data.\n\n## Resources (direct URI lookup)\n\nFive resources cover the same content as the tools, addressable by URI for clients that prefer the `resources/read`\nflow.\n\n| URI                        | Returns                                       |\n| -------------------------- | --------------------------------------------- |\n| `anc://registry`           | full denormalized catalog (concrete resource) |\n| `anc://tool/{slug}`        | single CLI record                             |\n| `anc://principle/{n}`      | single principle body and requirements        |\n| `anc://spec/{section}`     | single spec section body                      |\n| `anc://scorecard/{binary}` | cached scorecard for a CLI by binary name     |\n\nPer-item records live behind templates surfaced via `resources/templates/list`. `resources/list` returns only the one\nconcrete resource (`anc://registry`).\n\n## When things fail\n\nTwo error layers. The discriminator is whether the JSON-RPC envelope itself succeeded.\n\n| Symptom                                                | Layer        | Recovery                                                                                                                       |\n| ------------------------------------------------------ | ------------ | ------------------------------------------------------------------------------------------------------------------------------ |\n| `CallToolResult` with `isError: true`                  | Tool-level   | Read the text content; the message names the failure (validator rejection, infrastructure, rate-limit).                        |\n| `error.code: -32700` at HTTP 400                       | Transport    | Malformed JSON body; the envelope's `id` is `null` because the request never parsed. Fix the request and resend.               |\n| JSON-RPC envelope with `error.code: -32099`            | Transport    | Rate limit. Back off per the policy below; either limiter can trip this.                                                       |\n| `error.code: -32022` at HTTP 200, no `data.requested`  | Transport    | The legacy lane is disabled. `error.data.supported` lists the served revision (`2026-07-28`); resend as a modern request.      |\n| `error.code: -32022` at HTTP 400 with `data.requested` | Transport    | Your claimed protocol version is unsupported…

{"content":[{"text":"=== about ===\n---\ntitle: About this standard\ndescription: The agent-native CLI standard is a specification for command-line tools that behave predictably under agent control. Eight principles, enforced by RFC 2119 requirement tiers…\nurl: https://anc.dev/about\n---\n\n# About this standard\n\nThe agent-native CLI standard is a specification for command-line tools that behave predictably under agent control.\nEight principles, enforced by RFC 2119 requirement tiers (MUST / SHOULD / MAY), and measured by a companion linter\n([`anc`](https://anc.dev/audit)) that scores any CLI against them.\n\n## Provenance\n\nThis spec is authored and maintained in the open by Brett Davies, with contributions accepted via the channels below. It\nis a proposal pressure-tested in public, not a ratified industry standard. The goal is to converge on something worth\nratifying, by writing it down concretely first and inviting people to break it.\n\n## Prior art\n\nThe eight-principle structure draws on two distinct lineages.\n\n**Standards and methodologies that shaped the format:**\n\n- [Command Line Interface Guidelines (clig.dev)](https://clig.dev/): the closest direct prior art for CLI design\n  guidance; the Unix-philosophy distillation this spec departs from when agents change the human-only assumptions.\n- [The Twelve-Factor App (12factor.net)](https://12factor.net/): the numbered-principle layout, environment-first\n  configuration, and the discipline of writing each factor down concretely.\n- [IETF RFC 2119](https://www.rfc-editor.org/rfc/rfc2119): the MUST / SHOULD / MAY contract that turns prose into a\n  conformance bar.\n\n**Writing that informed the principles directly:**\n\n- Cloudflare's [Building a CLI for all of Cloudflare](https://blog.cloudflare.com/cf-cli-local-explorer) (2026-04-13)\n  and the [HN discussion that followed](https://news.ycombinator.com/item?id=47753689): the most concrete public\n  statement of \"agents are the primary customer\" of CLI tools, paired with crowdsourced failure modes from people who\n  run agents against CLIs every day. Quotes from that thread shaped the framing of P3 (progressive help), P4 (actionable\n  errors), and P6 (composable structure) directly. P2 and P6 mirror the Cloudflare team's own rules (`get` not `info`,\n  `--json` always, `--force` not `--skip-confirmations`).\n\nIf a specific principle's framing seems to echo prior writing, it probably does. Credit accrues to the people whose\npublic reasoning informed it; mistakes are mine.\n\n## Versioning\n\nThe spec uses semver-adjacent versioning with three tiers: MAJOR changes break citation IDs or remove MUSTs; MINOR\nchanges add MUSTs or promote SHOULDs; PATCH changes edit prose without shifting requirements. The current version\nappears in the footer of every page.\n\nPrinciple anchor slugs (`#p1-non-interactive-by-default` through `#p8-discoverable-skill-bundle`) are permanent. If a\nprinciple merges or splits in a future MAJOR version, the old slug will resolve as a permanent redirect to wherever the\nrequirement now lives; citations made today will not 404 after a future restructuring.\n\n## RFC 2119\n\nMUST, SHOULD, MAY, MUST NOT, and SHOULD NOT on this site carry their [RFC 2119](https://www.rfc-editor.org/rfc/rfc2119)\nmeanings. They are not emphatic; they are contractual. A tool that does not satisfy a MUST is non-conformant with the\nversion of the standard it claims to target.\n\n## License\n=== badge ===\n---\ntitle: Agent-native badge\ndescription: The agent-native badge is a small SVG that lets a CLI author advertise agent-native conformance on their tool's README. It links to the live scorecard for that tool, so any…\nurl: https://anc.dev/badge\n---\n\n# Agent-native badge\n\nThe agent-native badge is a small SVG that lets a CLI author advertise agent-native conformance on their tool's README.\nIt links to the live scorecard for that tool, so any reader can follow the claim to current evidence rather than trust a\nself-declaration.\n\nThis page is the convention. It defines what claiming the badge means, what it does not mean, and how the URL behaves\nwhen a tool's score changes.\n\n## What the badge looks like\n\nThe badge is rendered at build time and served as a static SVG. The site does not use shields.io, has no third-party\nrender dependency, and requires no account. The renderer is [`badge-maker`](https://www.npmjs.com/package/badge-maker),\nthe same library shields.io uses internally. Visually identical output, fully self-hosted.\n\nFor a tool named `<tool>`:\n\n```text\nhttps://anc.dev/badge/<tool>.svg\n```\n\nThe label reads `agent-native vMAJOR.MINOR` (the spec version the score is rooted in); the message is the rounded\npercent score; the color tracks the cohort band the score falls into (see [Color bands](#color-bands) below).\n\n## How to embed it\n\n```markdown\n[![agent-native](https://anc.dev/badge/<tool>.svg)](https://anc.dev/score/<tool>)\n```\n\nReplace `<tool>` with the tool's slug: the same name that appears on the leaderboard and in the registry. The badge\nlinks to the tool's per-tool scorecard page so a reader who clicks lands on the live evidence.\n\nPer-tool scorecard pages whose tool clears the eligibility floor render this snippet inline, ready to copy.\n","type":"text"}]}

cd /tmp; echo "=== ai.txt ==="; curl -sS -m 15 https://anc.dev/.well-known/ai.txt; echo "=== license tail of about ==="; curl -sS -m 15 https://anc.dev/about.md | tail -20; echo "=== contribute ==="; curl -sS -m 15 https://anc.dev/contribute.md | head -40

{"content":[{"text":"=== ai.txt ===\n# ai.txt for anc.dev\n# Declares AI-training and agent-access posture. Format may evolve as the\n# ai.txt convention ratifies; this file is the canonical statement.\n\nUser-Agent: *\nAllow: /\nAllow-AI-Training: yes\nAllow-Inference: yes\nProgrammatic-API: https://anc.dev/mcp\nContact: mailto:97-boss-beetle@icloud.com\n=== license tail of about ===\n\nPrinciple anchor slugs (`#p1-non-interactive-by-default` through `#p8-discoverable-skill-bundle`) are permanent. If a\nprinciple merges or splits in a future MAJOR version, the old slug will resolve as a permanent redirect to wherever the\nrequirement now lives; citations made today will not 404 after a future restructuring.\n\n## RFC 2119\n\nMUST, SHOULD, MAY, MUST NOT, and SHOULD NOT on this site carry their [RFC 2119](https://www.rfc-editor.org/rfc/rfc2119)\nmeanings. They are not emphatic; they are contractual. A tool that does not satisfy a MUST is non-conformant with the\nversion of the standard it claims to target.\n\n## License\n\nSpec text on this site is available under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). The `anc` linter is\ndual-licensed under MIT and Apache-2.0; see [its LICENSE files](https://github.com/brettdavies/agentnative-cli).\n\n## Colophon\n\nBuilt with Cloudflare Workers. Typeset in Uncut Sans and Monaspace Xenon. Source:\n[github.com/brettdavies/agentnative-site](https://github.com/brettdavies/agentnative-site).\n=== contribute ===\n---\ntitle: Contribute\ndescription: 'The agent-native CLI spec is `status: active` because the contracts are stable enough to cite, not because anything is locked. The pressure-test mechanism is how the spec…'\nurl: https://anc.dev/contribute\n---\n\n# Contribute\n\nThe agent-native CLI spec is `status: active` because the contracts are stable enough to cite, not because anything is\nlocked. The pressure-test mechanism is how the spec revises a position when a finding warrants it. This page is the\nnavigation across the four repos that make up the project, plus the honest expectations on response time.\n\n## What kinds of contribution are welcome\n\nThree tiers, all welcome, none required. The shape of the contribution determines the intake.\n\n| Tier            | What                                                                                                                                                                                                         | Where                                                              | Time     |\n| --------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | ------------------------------------------------------------------ | -------- |\n| **1. Signal**   | A finding against a principle's wording, a missing citation, a contradiction between two principles, a false positive in `anc`, a broken link on the site, a bundle content issue                            | A repo-specific issue template (see \"Per-repo intake\" below)       | ~5 min   |\n| **2. Proposal** | A new principle the spec is missing, a MUST/SHOULD tier change with rationale, a counter-example that breaks an applicability clause, a new language auditor design, a new host runtime for the skill bundle | An issue with the full case in the body, against the relevant repo | ~1-2 hrs |\n| **3. Code**     | A new language auditor for `anc`, a tool scoring submission for the leaderboard, a site or skill-bundle improvement, a governance or workflow PR                                                             | A pull request against the relevant repo's `dev` branch            | Variable |\n\n## Per-repo intake\n\nEach repo handles a different layer of the project. File against the one that matches the contribution's shape.\n\n### Spec: [agentnative](https://github.com/brettdavies/agentnative)\n\nThe principle text, the requirement IDs, the versioning policy. Pressure-tests against the standard live here.\n\n- [Pressure-test a principle](https://github.com/brettdavies/agentnative/issues/new?template=pressure-test.yml) (Tier 1\n  or 2)\n- [Ask a spec question](https://github.com/brettdavies/agentnative/issues/new?template=spec-question.yml) (Tier 1)\n- [Submit a grading finding](https://github.com/brettdavies/agentnative/issues/new?template=grading-finding.yml) (Tier 1\n  or 2): spec-feedback derived from scoring a real CLI against the standard\n- [`CONTRIBUTING.md`](https://github.com/brettdavies/agentnative/blob/main/CONTRIBUTING.md) ·\n  [`principles/AGENTS.md` § Pressure-test protocol](https://github.com/brettdavies/agentnative/blob/main/principles/AGENTS.md#pressure-test-protocol)\n\n### Linter: [agentnative-cli](https://github.com/brettdavies/agentnative-cli)\n\n","type":"text"}]}

cd /tmp; curl -sS -m 20 https://anc.dev/mcp -H 'Content-Type: application/json' -H 'Accept: application/json' -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}' | head -c 2000; echo; echo "--- modern ---"; curl -sS -m 20 https://anc.dev/mcp -H 'Content-Type: application/json' -H 'Accept: application/json' -H 'MCP-Protocol-Version: 2026-07-28' -H 'Mcp-Method: tools/list' -d '{"jsonrpc":"2.0","id":10,"method":"tools/list","params":{"_meta":{"io.modelcontextprotocol/protocolVersion":"2026-07-28","io.modelcontextprotocol/clientInfo":{"name":"demo","version":"0"},"io.modelcontextprotocol/clientCapabilities":{}}}}' | head -c 1500

{"content":[{"text":"{\"result\":{\"tools\":[{\"name\":\"list_tools\",\"title\":\"List scored CLI registry entries\",\"description\":\"Return summaries of every scored CLI in the anc.dev registry. Each entry carries slug, name, binary, install command, version (when a scorecard has been committed), score_pct (percentage pass rate of the latest audit), scorecard_url (path under anc.dev), and audit_profile (when the tool opts out of the default profile).\",\"inputSchema\":{\"type\":\"object\",\"properties\":{}},\"annotations\":{\"readOnlyHint\":true}},{\"name\":\"get_tool\",\"title\":\"Get a registry entry\",\"description\":\"Return the full registry record for a single CLI by slug. Fields: slug, name, binary, install, audit_profile (when set), repo (when a GitHub owner/repo is parseable), version, anc_version, scorecard_url, score_pct. Look-not-found returns isError: false with a typed { found: false, message } body because absence is data, not failure.\",\"inputSchema\":{\"type\":\"object\",\"$schema\":\"https://json-schema.org/draft/2020-12/schema\",\"properties\":{\"slug\":{\"type\":\"string\",\"description\":\"The CLI slug, e.g. \\\"ripgrep\\\" or \\\"curl\\\".\"}},\"required\":[\"slug\"]},\"annotations\":{\"readOnlyHint\":true}},{\"name\":\"search_tools\",\"title\":\"Search the CLI registry\",\"description\":\"Filter the registry by one or more criteria. All filters AND together. score_min / score_max are inclusive bounds on score_pct (rows without a committed scorecard are excluded when either bound is set). audit_profile is an exact match. principle_min_score is reserved for a future per-principle score filter and is currently a no-op (the catalog projection does not yet carry per-principle scores).\",\"inputSchema\":{\"type\":\"object\",\"$schema\":\"https://json-schema.org/draft/2020-12/schema\",\"properties\":{\"score_min\":{\"description\":\"Inclusive lower bound on score_pct.\",\"type\":\"number\",\"minimum\":0,\"maximum\":100},\"score_max\":{\"description\":\"Inclusive upper bound on score_pct.\",\"type\":\"number\",\"minimum\":0,\"maximum\":100},\"audit_profile\":{\"description\":\"Exact audit_profile matccurl: (23) Failure writing output to destination\n\n--- modern ---\n{\"result\":{\"tools\":[{\"name\":\"list_tools\",\"title\":\"List scored CLI registry entries\",\"description\":\"Return summaries of every scored CLI in the anc.dev registry. Each entry carries slug, name, binary, install command, version (when a scorecard has been committed), score_pct (percentage pass rate of the latest audit), scorecard_url (path under anc.dev), and audit_profile (when the tool opts out of the default profile).\",\"inputSchema\":{\"type\":\"object\",\"properties\":{}},\"annotations\":{\"readOnlyHint\":true}},{\"name\":\"get_tool\",\"title\":\"Get a registry entry\",\"description\":\"Return the full registry record for a single CLI by slug. Fields: slug, name, binary, install, audit_profile (when set), repo (when a GitHub owner/repo is parseable), version, anc_version, scorecard_url, score_pct. Look-not-found returns isError: false with a typed { found: false, message } body because absence is data, not failure.\",\"inputSchema\":{\"type\":\"object\",\"$schema\":\"https://json-schema.org/draft/2020-12/schema\",\"properties\":{\"slug\":{\"type\":\"string\",\"description\":\"The CLI slug, e.g. \\\"ripgrep\\\" or \\\"curl\\\".\"}},\"required\":[\"slug\"]},\"annotations\":{\"readOnlyHint\":true}},{\"name\":\"search_tools\",\"title\":\"Search the CLI registry\",\"description\":\"Filter the registry by one or more criteria. All filters AND together. score_min / score_max are inclusive bounds on score_pct (rows without a committed scorecard are excluded when either bound is set). audit_profile is an exact match. principle_min_score is reserved for a ","type":"text"}]}

cd /tmp; python3 -V; curl -sS -m 25 https://anc.dev/mcp -H 'Content-Type: application/json' -H 'Accept: application/json' -d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"get_scorecard","arguments":{"slug":"ripgrep"}}}' > sc.json; python3 -c "
import json;d=json.load(open('sc.json'));r=d['result'];print(r.get('isError'));t=r['content'][0]['text'];print(t[:1200])"

{"content":[{"text":"Python 3.12.3\nNone\n{\n  \"found\": true,\n  \"kind\": \"cli\",\n  \"tier\": \"registry\",\n  \"target\": \"ripgrep\",\n  \"scorecard_url\": \"https://anc.dev/score/ripgrep\",\n  \"markdown_url\": \"https://anc.dev/score/ripgrep/md\",\n  \"json_url\": \"https://anc.dev/score/ripgrep/json\",\n  \"freshness\": {\n    \"cached\": true,\n    \"scored_at\": null,\n    \"refresh_after\": null\n  },\n  \"spec_version\": \"0.5.0\",\n  \"scorecard\": {\n    \"schema_version\": \"0.7\",\n    \"results\": [\n      {\n        \"id\": \"p3-must-top-level-examples\",\n        \"label\": \"Help flag produces useful output\",\n        \"group\": \"P3\",\n        \"layer\": \"behavioral\",\n        \"status\": \"pass\",\n        \"evidence\": null,\n        \"confidence\": \"high\",\n        \"tier\": \"must\",\n        \"audit_id\": \"p3-help\"\n      },\n      {\n        \"id\": \"p3-must-version\",\n        \"label\": \"Version flag works (`--version` plus short alias)\",\n        \"group\": \"P3\",\n        \"layer\": \"behavioral\",\n        \"status\": \"pass\",\n        \"evidence\": null,\n        \"confidence\": \"high\",\n        \"tier\": \"must\",\n        \"audit_id\": \"p3-version\"\n      },\n      {\n        \"id\": \"p3-should-version-short\",\n        \"label\": \"Version flag works (`--version` plus short alias)\",\n        \"group\": \"P3\",\n        \"layer\": \"beh\n","type":"text"}]}

{"content":[{"text":"\nblocked: HTTP 403 from https://anc.dev/mcp: {\"type\":\"https://developers.cloudflare.com/support/troubleshooting/http-status-codes/cloudflare-1xxx-errors/error-1010/\",\"title\":\"Error 1010: Access denied\",\"status\":403,\"detail\":\"The site owner has blocked access based on your browser's signature.\",\"instance\":\"a3d36a9d9aaafa75\",\"error_code\":1010,\"error_name\":\"browser_signature_banned\",\"error_category\":\"access_denied\",\"ray_id\":\"a3d36a9d9aaafa75\",\"\n","type":"text"}]}

{"content":[{"text":"Successfully replaced 2 block(s) in /sandbox/repo/anc_mcp.py.","type":"text"}],"details":{"diff":"     ...\n  15 from typing import Any\n  16 \n  17 ENDPOINT = \"https://anc.dev/mcp\"\n  18 PROTOCOL_VERSION = \"2026-07-28\"\n+ 19 USER_AGENT = \"anc-demo/0.1 (mcp client; +https://anc.dev/mcp-skill)\"\n  19 \n  20 # Cheap read tools (registry / cache lookups): 60 req / 60 s per IP.\n  21 # Metered tools (fresh container audit / web audit): much tighter -- see README.\n  22 READ_TOOLS = {\n     ...\n  45         data=json.dumps(payload).encode(),\n  46         headers={\n  47             \"Content-Type\": \"application/json\",\n  48             \"Accept\": \"application/json\",\n+ 50             # Cloudflare error 1010 blocks the default Python-urllib signature.\n+ 51             \"User-Agent\": USER_AGENT,\n  49             # Modern-lane headers; the server accepts these alongside the body.\n  50             \"MCP-Protocol-Version\": PROTOCOL_VERSION,\n  51             \"Mcp-Method\": payload[\"method\"],\n  52             **({\"Mcp-Name\": payload[\"params\"][\"name\"]} if \"name\" in payload.get(\"params\", {}) else {}),\n     ...","patch":"--- /sandbox/repo/anc_mcp.py\n+++ /sandbox/repo/anc_mcp.py\n@@ -15,8 +15,9 @@\n from typing import Any\n \n ENDPOINT = \"https://anc.dev/mcp\"\n PROTOCOL_VERSION = \"2026-07-28\"\n+USER_AGENT = \"anc-demo/0.1 (mcp client; +https://anc.dev/mcp-skill)\"\n \n # Cheap read tools (registry / cache lookups): 60 req / 60 s per IP.\n # Metered tools (fresh container audit / web audit): much tighter -- see README.\n READ_TOOLS = {\n@@ -45,8 +46,10 @@\n         data=json.dumps(payload).encode(),\n         headers={\n             \"Content-Type\": \"application/json\",\n             \"Accept\": \"application/json\",\n+            # Cloudflare error 1010 blocks the default Python-urllib signature.\n+            \"User-Agent\": USER_AGENT,\n             # Modern-lane headers; the server accepts these alongside the body.\n             \"MCP-Protocol-Version\": PROTOCOL_VERSION,\n             \"Mcp-Method\": payload[\"method\"],\n             **({\"Mcp-Name\": payload[\"params\"][\"name\"]} if \"name\" in payload.get(\"params\", {}) else {}),\n","firstChangedLine":19}}

{"content":[{"text":"\n=== ripgrep :: https://anc.dev/score/ripgrep\ntier=registry  spec=0.5.0  score=82%\nstatuses: {'pass': 20, 'opt_out': 1, 'skip': 14, 'warn': 6, 'n_a': 1, 'fail': 1}\n\nnon-passing checks (23):\n  [     opt_out] MUST  p2-must-output-flag - Structured output support\n  [        skip] MUST  p1-must-no-interactive - Non-interactive gate flag advertised in --help\n  [        warn] MUST  p6-must-no-pager - Pager-using CLI ships --no-pager escape hatch\n  [         n_a] MUST  p2-must-schema-print - Structured-output CLI exposes its schema at runtime\n  [        skip] MAY   p6-may-standard-names - Subcommand verbs follow community-standard names\n  [        warn] MAY   p2-may-raw-flag - `--raw` flag for pipe-safe unformatted output\n  [        skip] MAY   p2-may-more-formats - `--output` advertises additional formats beyond text/json\n  [        warn] MAY   p3-may-examples-subcommand - `examples` subcommand or `--examples` flag for curated usage patterns\n  [        skip] SHOULD p7-should-limit - `--limit` / `--max-results` flag for list operations\n  [        skip] MAY   p7-may-cursor-pagination - Cursor-based pagination flags for list traversal\n  ... and 13 more\n\n=== registry: 98 entries, top 5 by score\n  100%  anc              brew install brettdavies/tap/agentnative\n   88%  watchexec        brew install watchexec\n   85%  miniserve        brew install miniserve\n   84%  bat              brew install bat\n   84%  delta            brew install git-delta\n\n=== P1: P1: Non-Interactive by Default\n  [MUST  ] p1-must-env-var: Every flag settable via environment variable (falsey-value parser for booleans).\n  [MUST  ] p1-must-no-interactive: When stdin is not a TTY or `--no-interactive` is set, every blocking-input surface (prompt libraries, read-line, TUI init) resolves from defaults/stdin or exits with an actionable error.\n  [MUST  ] p1-must-no-browser: Headless authentication path (`--no-browser` / OAuth Device Authorization Grant).\n  [MUST  ] p1-must-secret-non-leaky-path: Sensitive inputs are readable via stdin or a `--*-file` flag; flag-value and env-var inputs MAY exist for convenience but MUST NOT be the only path.\n  [SHOULD] p1-should-tty-detection: Auto-detect non-interactive context via TTY detection; suppress prompts when stderr is not a terminal.\n  [SHOULD] p1-should-defaults-in-help: Document default values for prompted inputs in `--help` output.\n","type":"text"}]}

{"content":[{"text":"Successfully replaced 1 block(s) in /sandbox/repo/demo.py.","type":"text"}],"details":{"diff":"    ...\n 51     if not env.get(\"found\"):\n 52         print(f\"no principle {n}\")\n 53         return\n 54     p = env[\"principle\"]\n-55     print(f\"\\n=== P{p['n']}: {p['title']}\")\n+55     title = p[\"title\"].removeprefix(f\"P{p['n']}: \")\n+56     print(f\"\\n=== P{p['n']}: {title}\")\n 56     for req in p.get(\"requirements\", [])[:6]:\n 57         print(f\"  [{req['level'].upper():6}] {req['id']}: {req['summary']}\")\n 58 \n 59 \n    ...","patch":"--- /sandbox/repo/demo.py\n+++ /sandbox/repo/demo.py\n@@ -51,9 +51,10 @@\n     if not env.get(\"found\"):\n         print(f\"no principle {n}\")\n         return\n     p = env[\"principle\"]\n-    print(f\"\\n=== P{p['n']}: {p['title']}\")\n+    title = p[\"title\"].removeprefix(f\"P{p['n']}: \")\n+    print(f\"\\n=== P{p['n']}: {title}\")\n     for req in p.get(\"requirements\", [])[:6]:\n         print(f\"  [{req['level'].upper():6}] {req['id']}: {req['summary']}\")\n \n \n","firstChangedLine":55}}

{"content":[{"text":"  ... and 13 more\n\n=== registry: 98 entries, top 5 by score\n  100%  anc              brew install brettdavies/tap/agentnative\n   88%  watchexec        brew install watchexec\n   85%  miniserve        brew install miniserve\n   84%  bat              brew install bat\n   84%  delta            brew install git-delta\n\n=== P3: Progressive Help Discovery\n  [MUST  ] p3-must-subcommand-examples: Every subcommand ships at least one concrete invocation example (`after_help` in clap).\n  [MUST  ] p3-must-top-level-examples: The top-level command ships 2–3 examples covering the primary use cases.\n  [MUST  ] p3-must-version: Top-level `--version` prints a non-empty version line and exits 0.\n  [SHOULD] p3-should-version-short: A short version alias (`-V`, `-v`, or `-version`) accompanies `--version` for fast version probes.\n  [SHOULD] p3-should-paired-examples: Examples show human and agent invocations side by side (text then `--output json` equivalent).\n  [SHOULD] p3-should-about-long-about: Short `about` for command-list summaries; `long_about` reserved for detailed descriptions visible with `--help`.\n\n=== web audit: anc.dev\n  score={'relative': 100, 'global': 84}  https://anc.dev/score/anc.dev\n  [n_a] openapi - An OpenAPI description is published\n","type":"text"}]}

cd /tmp; curl -sS -m 15 https://anc.dev/sitemap.xml | grep -o '<loc>[^<]*' | sed 's/<loc>//' | grep -vi '/score/' ; echo "--- grep pricing words"; for p in about.md methodology.md changelog.md mcp-skill.md; do echo "[$p]"; curl -sS -m 15 https://anc.dev/$p | grep -inE 'pric|cost|free|paid|plan|billing|api key|auth' | head -8; done

{"content":[{"text":"https://anc.dev/\nhttps://anc.dev/p1\nhttps://anc.dev/p2\nhttps://anc.dev/p3\nhttps://anc.dev/p4\nhttps://anc.dev/p5\nhttps://anc.dev/p6\nhttps://anc.dev/p7\nhttps://anc.dev/p8\nhttps://anc.dev/audit\nhttps://anc.dev/about\nhttps://anc.dev/changelog\nhttps://anc.dev/contribute\nhttps://anc.dev/methodology\nhttps://anc.dev/scorecard-schema\nhttps://anc.dev/mcp-skill\nhttps://anc.dev/mcp\nhttps://anc.dev/scorecards\nhttps://anc.dev/coverage\nhttps://anc.dev/install\nhttps://anc.dev/skill\nhttps://anc.dev/badge\nhttps://anc.dev/web-scorecard-schema\n--- grep pricing words\n[about.md]\n15:This spec is authored and maintained in the open by Brett Davies, with contributions accepted via the channels below. It\n[methodology.md]\n124:decides *how much* a miss costs once it does.\n166:The classifier is **informational, not authoritative**. It is a one-line summary derived from a fixed set of four\n335:the requirement and the fix guidance. The leaderboard exists to make the standard concrete, not to shame tool authors\n[changelog.md]\n[mcp-skill.md]\n11:lookup. The catalog is public: no authentication, no API key.\n68:The most-used surface. Two tools, three input shapes, one orchestration core. The split is honest about cost:\n95:Use `tier` as your cost signal: `registry` means curated and committed, `cache` means a prior audit cached the result,\n144:container run, no cost). On a true cache miss it runs a metered audit and returns the fresh scorecard.\n186:same input. The cost difference (registry/cache lookup vs container run) is the only reason to choose between them.\n410:Two limiters, two cost profiles.\n418:container-run cost is non-trivial and a shared anon bucket would be a DoS vector. The hourly ceiling is enforced in two\n431:`Allow: GET, POST`. No authentication.\n","type":"text"}]}