@rmyndharis/aimhooman 0.1.0

package/.agents/rules/aimhooman.md

@@ -0,0 +1,53 @@
+# aimhooman
+
+<!-- aimhooman:ruleset-start -->
+This repository uses aimhooman to keep AI tooling artifacts out of Git history.
+Its policy:
+
+- Never stage or commit local AI session/state files. Examples include
+  `.claude.json`, `.claude/.credentials.json`, `.claude/session*.json`,
+  `.claude/history*`, `.claude/todos/*`, `.claude/shell-snapshots/*`,
+  `.claude/statsig/*`, `.claude/projects/*`, `.claude/logs/*`,
+  `.codex/sessions/*`, `.codex/history*`, `.codex/log/*`, `.codex/logs/*`,
+  `.copilot/*`, `.cursor/session*`, `.cursor/chats/*`, `.cursor/logs/*`,
+  `.aider.*`, `.specstory/*`, `.continue/sessions/*`, `.playwright-mcp/*`,
+  `.remember/*`, `.superpowers/*`, and `.agent/*`. The examples are not
+  exhaustive; the packaged `rules/paths.json` is the detection source of truth.
+  If one is staged, unstage it and keep it out of Git instead.
+- Never commit secrets: a real `.env` (not `.env.example`), private keys
+  (`id_rsa` and files containing a private-key header), `.aws/credentials`, or
+  service-account keys. Public certificates are allowed.
+- Never add AI attribution to commit messages: no `Co-authored-by` trailer naming
+  an AI (Claude, Copilot, Codex), no "Generated with/by <AI>" lines, no AI-service
+  noreply emails. A commit message reads as if a human wrote it.
+- Write in a human voice, not an AI voice. No `delve`, `leverage`, `utilize`,
+  `seamless`, `robust`, `comprehensive`, `underscore`, `foster`, `pivotal`, `myriad`,
+  `tapestry`, `landscape`; no "This PR addresses…", "adversarial self-review",
+  "live-verified", "thoroughly tested with comprehensive examples", "in today's",
+  "let's dive in", "in conclusion"; no bold lead-in on every bullet and no paired
+  em-dash asides. Say what broke or what changed, pick the plain short verb, name the
+  number, and vary sentence length. If the first line could open any PR on any repo,
+  delete it.
+- Treat `AGENTS.md`, `CLAUDE.md`, `GEMINI.md`, and `.github/copilot-instructions.md`
+  as review-required: they may be intentional team config, so confirm before
+  committing them.
+
+Project map:
+
+- `src/scan.mjs` and `src/rules.mjs` load rules and make policy decisions.
+- `src/hook.mjs` and `src/githooks.mjs` enforce them in agent and Git hooks.
+- `src/state.mjs` owns local policy state. `bin/aimhooman.mjs` is the CLI.
+- `tests/` mirrors the production paths and includes real temporary Git repos.
+
+Use Node.js 22.8 or newer. Before finishing a change, run:
+
+```sh
+npm run check
+npm test
+npm run test:coverage
+```
+
+aimhooman also enforces this at commit time, so a blocked commit means a real
+violation to fix, not a check to bypass. The rule is simple: AI works, hoomans
+ship.
+<!-- aimhooman:ruleset-end -->

package/.claude-plugin/marketplace.json

@@ -0,0 +1,23 @@
+{
+    "name": "aimhooman",
+    "owner": {
+        "name": "aimhooman"
+    },
+    "description": "AI works. Hoomans ship. Keep AI session files, secrets, and attribution out of your commits.",
+    "plugins": [
+        {
+            "name": "aimhooman",
+            "source": "./",
+            "description": "Keeps AI session files, secrets, and unwanted attribution out at the agent and Git boundary.",
+            "version": "0.1.0",
+            "license": "MIT",
+            "keywords": [
+                "git",
+                "hooks",
+                "ai-agents",
+                "security",
+                "developer-tools"
+            ]
+        }
+    ]
+}

package/.claude-plugin/plugin.json

@@ -0,0 +1,9 @@
+{
+    "name": "aimhooman",
+    "version": "0.1.0",
+    "description": "AI works. Hoomans ship. Keeps AI session files, secrets, and unwanted attribution out at the agent and Git boundary.",
+    "author": {
+        "name": "aimhooman"
+    },
+    "hooks": "./hooks/hooks.json"
+}

package/.clinerules/aimhooman.md

@@ -0,0 +1,53 @@
+# aimhooman
+
+<!-- aimhooman:ruleset-start -->
+This repository uses aimhooman to keep AI tooling artifacts out of Git history.
+Its policy:
+
+- Never stage or commit local AI session/state files. Examples include
+  `.claude.json`, `.claude/.credentials.json`, `.claude/session*.json`,
+  `.claude/history*`, `.claude/todos/*`, `.claude/shell-snapshots/*`,
+  `.claude/statsig/*`, `.claude/projects/*`, `.claude/logs/*`,
+  `.codex/sessions/*`, `.codex/history*`, `.codex/log/*`, `.codex/logs/*`,
+  `.copilot/*`, `.cursor/session*`, `.cursor/chats/*`, `.cursor/logs/*`,
+  `.aider.*`, `.specstory/*`, `.continue/sessions/*`, `.playwright-mcp/*`,
+  `.remember/*`, `.superpowers/*`, and `.agent/*`. The examples are not
+  exhaustive; the packaged `rules/paths.json` is the detection source of truth.
+  If one is staged, unstage it and keep it out of Git instead.
+- Never commit secrets: a real `.env` (not `.env.example`), private keys
+  (`id_rsa` and files containing a private-key header), `.aws/credentials`, or
+  service-account keys. Public certificates are allowed.
+- Never add AI attribution to commit messages: no `Co-authored-by` trailer naming
+  an AI (Claude, Copilot, Codex), no "Generated with/by <AI>" lines, no AI-service
+  noreply emails. A commit message reads as if a human wrote it.
+- Write in a human voice, not an AI voice. No `delve`, `leverage`, `utilize`,
+  `seamless`, `robust`, `comprehensive`, `underscore`, `foster`, `pivotal`, `myriad`,
+  `tapestry`, `landscape`; no "This PR addresses…", "adversarial self-review",
+  "live-verified", "thoroughly tested with comprehensive examples", "in today's",
+  "let's dive in", "in conclusion"; no bold lead-in on every bullet and no paired
+  em-dash asides. Say what broke or what changed, pick the plain short verb, name the
+  number, and vary sentence length. If the first line could open any PR on any repo,
+  delete it.
+- Treat `AGENTS.md`, `CLAUDE.md`, `GEMINI.md`, and `.github/copilot-instructions.md`
+  as review-required: they may be intentional team config, so confirm before
+  committing them.
+
+Project map:
+
+- `src/scan.mjs` and `src/rules.mjs` load rules and make policy decisions.
+- `src/hook.mjs` and `src/githooks.mjs` enforce them in agent and Git hooks.
+- `src/state.mjs` owns local policy state. `bin/aimhooman.mjs` is the CLI.
+- `tests/` mirrors the production paths and includes real temporary Git repos.
+
+Use Node.js 22.8 or newer. Before finishing a change, run:
+
+```sh
+npm run check
+npm test
+npm run test:coverage
+```
+
+aimhooman also enforces this at commit time, so a blocked commit means a real
+violation to fix, not a check to bypass. The rule is simple: AI works, hoomans
+ship.
+<!-- aimhooman:ruleset-end -->

package/.codex-plugin/plugin.json

@@ -0,0 +1,37 @@
+{
+    "name": "aimhooman",
+    "version": "0.1.0",
+    "description": "AI works. Hoomans ship. Keeps AI session files, secrets, and unwanted attribution out at the agent and Git boundary.",
+    "author": {
+        "name": "aimhooman maintainers",
+        "email": "yudhi@rmyndharis.com",
+        "url": "https://github.com/rmyndharis"
+    },
+    "homepage": "https://github.com/rmyndharis/aimhooman#readme",
+    "repository": "https://github.com/rmyndharis/aimhooman",
+    "license": "MIT",
+    "keywords": [
+        "ai-agents",
+        "git-hooks",
+        "repository-hygiene"
+    ],
+    "skills": "./skills/",
+    "interface": {
+        "displayName": "aimhooman",
+        "shortDescription": "Keep AI tooling residue out of Git",
+        "longDescription": "Detects AI session files, secrets, and unwanted attribution before they enter Git history.",
+        "developerName": "aimhooman maintainers",
+        "category": "Developer Tools",
+        "capabilities": [
+            "Interactive",
+            "Write"
+        ],
+        "websiteURL": "https://github.com/rmyndharis/aimhooman",
+        "defaultPrompt": [
+            "Check this repository for AI tooling artifacts.",
+            "Explain the aimhooman policy for this repository.",
+            "Audit the current Git range with aimhooman."
+        ],
+        "brandColor": "#111827"
+    }
+}

package/.cursor/rules/aimhooman.mdc

@@ -0,0 +1,59 @@
+---
+description: aimhooman - keep AI session files, secrets, and unwanted AI attribution out of Git.
+globs:
+alwaysApply: true
+---
+
+# aimhooman
+
+<!-- aimhooman:ruleset-start -->
+This repository uses aimhooman to keep AI tooling artifacts out of Git history.
+Its policy:
+
+- Never stage or commit local AI session/state files. Examples include
+  `.claude.json`, `.claude/.credentials.json`, `.claude/session*.json`,
+  `.claude/history*`, `.claude/todos/*`, `.claude/shell-snapshots/*`,
+  `.claude/statsig/*`, `.claude/projects/*`, `.claude/logs/*`,
+  `.codex/sessions/*`, `.codex/history*`, `.codex/log/*`, `.codex/logs/*`,
+  `.copilot/*`, `.cursor/session*`, `.cursor/chats/*`, `.cursor/logs/*`,
+  `.aider.*`, `.specstory/*`, `.continue/sessions/*`, `.playwright-mcp/*`,
+  `.remember/*`, `.superpowers/*`, and `.agent/*`. The examples are not
+  exhaustive; the packaged `rules/paths.json` is the detection source of truth.
+  If one is staged, unstage it and keep it out of Git instead.
+- Never commit secrets: a real `.env` (not `.env.example`), private keys
+  (`id_rsa` and files containing a private-key header), `.aws/credentials`, or
+  service-account keys. Public certificates are allowed.
+- Never add AI attribution to commit messages: no `Co-authored-by` trailer naming
+  an AI (Claude, Copilot, Codex), no "Generated with/by <AI>" lines, no AI-service
+  noreply emails. A commit message reads as if a human wrote it.
+- Write in a human voice, not an AI voice. No `delve`, `leverage`, `utilize`,
+  `seamless`, `robust`, `comprehensive`, `underscore`, `foster`, `pivotal`, `myriad`,
+  `tapestry`, `landscape`; no "This PR addresses…", "adversarial self-review",
+  "live-verified", "thoroughly tested with comprehensive examples", "in today's",
+  "let's dive in", "in conclusion"; no bold lead-in on every bullet and no paired
+  em-dash asides. Say what broke or what changed, pick the plain short verb, name the
+  number, and vary sentence length. If the first line could open any PR on any repo,
+  delete it.
+- Treat `AGENTS.md`, `CLAUDE.md`, `GEMINI.md`, and `.github/copilot-instructions.md`
+  as review-required: they may be intentional team config, so confirm before
+  committing them.
+
+Project map:
+
+- `src/scan.mjs` and `src/rules.mjs` load rules and make policy decisions.
+- `src/hook.mjs` and `src/githooks.mjs` enforce them in agent and Git hooks.
+- `src/state.mjs` owns local policy state. `bin/aimhooman.mjs` is the CLI.
+- `tests/` mirrors the production paths and includes real temporary Git repos.
+
+Use Node.js 22.8 or newer. Before finishing a change, run:
+
+```sh
+npm run check
+npm test
+npm run test:coverage
+```
+
+aimhooman also enforces this at commit time, so a blocked commit means a real
+violation to fix, not a check to bypass. The rule is simple: AI works, hoomans
+ship.
+<!-- aimhooman:ruleset-end -->

package/.gemini/settings.json

@@ -0,0 +1,7 @@
+{
+    "context": {
+        "fileName": [
+            "AGENTS.md"
+        ]
+    }
+}

package/.github/copilot-instructions.md

@@ -0,0 +1,53 @@
+# aimhooman
+
+<!-- aimhooman:ruleset-start -->
+This repository uses aimhooman to keep AI tooling artifacts out of Git history.
+Its policy:
+
+- Never stage or commit local AI session/state files. Examples include
+  `.claude.json`, `.claude/.credentials.json`, `.claude/session*.json`,
+  `.claude/history*`, `.claude/todos/*`, `.claude/shell-snapshots/*`,
+  `.claude/statsig/*`, `.claude/projects/*`, `.claude/logs/*`,
+  `.codex/sessions/*`, `.codex/history*`, `.codex/log/*`, `.codex/logs/*`,
+  `.copilot/*`, `.cursor/session*`, `.cursor/chats/*`, `.cursor/logs/*`,
+  `.aider.*`, `.specstory/*`, `.continue/sessions/*`, `.playwright-mcp/*`,
+  `.remember/*`, `.superpowers/*`, and `.agent/*`. The examples are not
+  exhaustive; the packaged `rules/paths.json` is the detection source of truth.
+  If one is staged, unstage it and keep it out of Git instead.
+- Never commit secrets: a real `.env` (not `.env.example`), private keys
+  (`id_rsa` and files containing a private-key header), `.aws/credentials`, or
+  service-account keys. Public certificates are allowed.
+- Never add AI attribution to commit messages: no `Co-authored-by` trailer naming
+  an AI (Claude, Copilot, Codex), no "Generated with/by <AI>" lines, no AI-service
+  noreply emails. A commit message reads as if a human wrote it.
+- Write in a human voice, not an AI voice. No `delve`, `leverage`, `utilize`,
+  `seamless`, `robust`, `comprehensive`, `underscore`, `foster`, `pivotal`, `myriad`,
+  `tapestry`, `landscape`; no "This PR addresses…", "adversarial self-review",
+  "live-verified", "thoroughly tested with comprehensive examples", "in today's",
+  "let's dive in", "in conclusion"; no bold lead-in on every bullet and no paired
+  em-dash asides. Say what broke or what changed, pick the plain short verb, name the
+  number, and vary sentence length. If the first line could open any PR on any repo,
+  delete it.
+- Treat `AGENTS.md`, `CLAUDE.md`, `GEMINI.md`, and `.github/copilot-instructions.md`
+  as review-required: they may be intentional team config, so confirm before
+  committing them.
+
+Project map:
+
+- `src/scan.mjs` and `src/rules.mjs` load rules and make policy decisions.
+- `src/hook.mjs` and `src/githooks.mjs` enforce them in agent and Git hooks.
+- `src/state.mjs` owns local policy state. `bin/aimhooman.mjs` is the CLI.
+- `tests/` mirrors the production paths and includes real temporary Git repos.
+
+Use Node.js 22.8 or newer. Before finishing a change, run:
+
+```sh
+npm run check
+npm test
+npm run test:coverage
+```
+
+aimhooman also enforces this at commit time, so a blocked commit means a real
+violation to fix, not a check to bypass. The rule is simple: AI works, hoomans
+ship.
+<!-- aimhooman:ruleset-end -->

package/.github/hooks/aimhooman.json

@@ -0,0 +1,22 @@
+{
+  "version": 1,
+  "hooks": {
+    "sessionStart": [
+      {
+        "type": "command",
+        "bash": "if command -v aimhooman >/dev/null 2>&1; then aimhooman hook session-start; else exit 0; fi",
+        "powershell": "if (Get-Command aimhooman -ErrorAction SilentlyContinue) { & aimhooman hook session-start; exit $LASTEXITCODE } else { exit 0 }",
+        "timeoutSec": 5
+      }
+    ],
+    "preToolUse": [
+      {
+        "type": "command",
+        "matcher": ".*",
+        "bash": "if command -v aimhooman >/dev/null 2>&1; then aimhooman hook pre-tool-use; else exit 0; fi",
+        "powershell": "if (Get-Command aimhooman -ErrorAction SilentlyContinue) { & aimhooman hook pre-tool-use; exit $LASTEXITCODE } else { exit 0 }",
+        "timeoutSec": 5
+      }
+    ]
+  }
+}

package/.kiro/steering/aimhooman.md

@@ -0,0 +1,53 @@
+# aimhooman
+
+<!-- aimhooman:ruleset-start -->
+This repository uses aimhooman to keep AI tooling artifacts out of Git history.
+Its policy:
+
+- Never stage or commit local AI session/state files. Examples include
+  `.claude.json`, `.claude/.credentials.json`, `.claude/session*.json`,
+  `.claude/history*`, `.claude/todos/*`, `.claude/shell-snapshots/*`,
+  `.claude/statsig/*`, `.claude/projects/*`, `.claude/logs/*`,
+  `.codex/sessions/*`, `.codex/history*`, `.codex/log/*`, `.codex/logs/*`,
+  `.copilot/*`, `.cursor/session*`, `.cursor/chats/*`, `.cursor/logs/*`,
+  `.aider.*`, `.specstory/*`, `.continue/sessions/*`, `.playwright-mcp/*`,
+  `.remember/*`, `.superpowers/*`, and `.agent/*`. The examples are not
+  exhaustive; the packaged `rules/paths.json` is the detection source of truth.
+  If one is staged, unstage it and keep it out of Git instead.
+- Never commit secrets: a real `.env` (not `.env.example`), private keys
+  (`id_rsa` and files containing a private-key header), `.aws/credentials`, or
+  service-account keys. Public certificates are allowed.
+- Never add AI attribution to commit messages: no `Co-authored-by` trailer naming
+  an AI (Claude, Copilot, Codex), no "Generated with/by <AI>" lines, no AI-service
+  noreply emails. A commit message reads as if a human wrote it.
+- Write in a human voice, not an AI voice. No `delve`, `leverage`, `utilize`,
+  `seamless`, `robust`, `comprehensive`, `underscore`, `foster`, `pivotal`, `myriad`,
+  `tapestry`, `landscape`; no "This PR addresses…", "adversarial self-review",
+  "live-verified", "thoroughly tested with comprehensive examples", "in today's",
+  "let's dive in", "in conclusion"; no bold lead-in on every bullet and no paired
+  em-dash asides. Say what broke or what changed, pick the plain short verb, name the
+  number, and vary sentence length. If the first line could open any PR on any repo,
+  delete it.
+- Treat `AGENTS.md`, `CLAUDE.md`, `GEMINI.md`, and `.github/copilot-instructions.md`
+  as review-required: they may be intentional team config, so confirm before
+  committing them.
+
+Project map:
+
+- `src/scan.mjs` and `src/rules.mjs` load rules and make policy decisions.
+- `src/hook.mjs` and `src/githooks.mjs` enforce them in agent and Git hooks.
+- `src/state.mjs` owns local policy state. `bin/aimhooman.mjs` is the CLI.
+- `tests/` mirrors the production paths and includes real temporary Git repos.
+
+Use Node.js 22.8 or newer. Before finishing a change, run:
+
+```sh
+npm run check
+npm test
+npm run test:coverage
+```
+
+aimhooman also enforces this at commit time, so a blocked commit means a real
+violation to fix, not a check to bypass. The rule is simple: AI works, hoomans
+ship.
+<!-- aimhooman:ruleset-end -->

package/.windsurf/rules/aimhooman.md

@@ -0,0 +1,57 @@
+---
+trigger: always_on
+---
+
+# aimhooman
+
+<!-- aimhooman:ruleset-start -->
+This repository uses aimhooman to keep AI tooling artifacts out of Git history.
+Its policy:
+
+- Never stage or commit local AI session/state files. Examples include
+  `.claude.json`, `.claude/.credentials.json`, `.claude/session*.json`,
+  `.claude/history*`, `.claude/todos/*`, `.claude/shell-snapshots/*`,
+  `.claude/statsig/*`, `.claude/projects/*`, `.claude/logs/*`,
+  `.codex/sessions/*`, `.codex/history*`, `.codex/log/*`, `.codex/logs/*`,
+  `.copilot/*`, `.cursor/session*`, `.cursor/chats/*`, `.cursor/logs/*`,
+  `.aider.*`, `.specstory/*`, `.continue/sessions/*`, `.playwright-mcp/*`,
+  `.remember/*`, `.superpowers/*`, and `.agent/*`. The examples are not
+  exhaustive; the packaged `rules/paths.json` is the detection source of truth.
+  If one is staged, unstage it and keep it out of Git instead.
+- Never commit secrets: a real `.env` (not `.env.example`), private keys
+  (`id_rsa` and files containing a private-key header), `.aws/credentials`, or
+  service-account keys. Public certificates are allowed.
+- Never add AI attribution to commit messages: no `Co-authored-by` trailer naming
+  an AI (Claude, Copilot, Codex), no "Generated with/by <AI>" lines, no AI-service
+  noreply emails. A commit message reads as if a human wrote it.
+- Write in a human voice, not an AI voice. No `delve`, `leverage`, `utilize`,
+  `seamless`, `robust`, `comprehensive`, `underscore`, `foster`, `pivotal`, `myriad`,
+  `tapestry`, `landscape`; no "This PR addresses…", "adversarial self-review",
+  "live-verified", "thoroughly tested with comprehensive examples", "in today's",
+  "let's dive in", "in conclusion"; no bold lead-in on every bullet and no paired
+  em-dash asides. Say what broke or what changed, pick the plain short verb, name the
+  number, and vary sentence length. If the first line could open any PR on any repo,
+  delete it.
+- Treat `AGENTS.md`, `CLAUDE.md`, `GEMINI.md`, and `.github/copilot-instructions.md`
+  as review-required: they may be intentional team config, so confirm before
+  committing them.
+
+Project map:
+
+- `src/scan.mjs` and `src/rules.mjs` load rules and make policy decisions.
+- `src/hook.mjs` and `src/githooks.mjs` enforce them in agent and Git hooks.
+- `src/state.mjs` owns local policy state. `bin/aimhooman.mjs` is the CLI.
+- `tests/` mirrors the production paths and includes real temporary Git repos.
+
+Use Node.js 22.8 or newer. Before finishing a change, run:
+
+```sh
+npm run check
+npm test
+npm run test:coverage
+```
+
+aimhooman also enforces this at commit time, so a blocked commit means a real
+violation to fix, not a check to bypass. The rule is simple: AI works, hoomans
+ship.
+<!-- aimhooman:ruleset-end -->

package/AGENTS.md

@@ -0,0 +1,53 @@
+# aimhooman: ship it like a hooman
+
+<!-- aimhooman:ruleset-start -->
+This repository uses aimhooman to keep AI tooling artifacts out of Git history.
+Its policy:
+
+- Never stage or commit local AI session/state files. Examples include
+  `.claude.json`, `.claude/.credentials.json`, `.claude/session*.json`,
+  `.claude/history*`, `.claude/todos/*`, `.claude/shell-snapshots/*`,
+  `.claude/statsig/*`, `.claude/projects/*`, `.claude/logs/*`,
+  `.codex/sessions/*`, `.codex/history*`, `.codex/log/*`, `.codex/logs/*`,
+  `.copilot/*`, `.cursor/session*`, `.cursor/chats/*`, `.cursor/logs/*`,
+  `.aider.*`, `.specstory/*`, `.continue/sessions/*`, `.playwright-mcp/*`,
+  `.remember/*`, `.superpowers/*`, and `.agent/*`. The examples are not
+  exhaustive; the packaged `rules/paths.json` is the detection source of truth.
+  If one is staged, unstage it and keep it out of Git instead.
+- Never commit secrets: a real `.env` (not `.env.example`), private keys
+  (`id_rsa` and files containing a private-key header), `.aws/credentials`, or
+  service-account keys. Public certificates are allowed.
+- Never add AI attribution to commit messages: no `Co-authored-by` trailer naming
+  an AI (Claude, Copilot, Codex), no "Generated with/by <AI>" lines, no AI-service
+  noreply emails. A commit message reads as if a human wrote it.
+- Write in a human voice, not an AI voice. No `delve`, `leverage`, `utilize`,
+  `seamless`, `robust`, `comprehensive`, `underscore`, `foster`, `pivotal`, `myriad`,
+  `tapestry`, `landscape`; no "This PR addresses…", "adversarial self-review",
+  "live-verified", "thoroughly tested with comprehensive examples", "in today's",
+  "let's dive in", "in conclusion"; no bold lead-in on every bullet and no paired
+  em-dash asides. Say what broke or what changed, pick the plain short verb, name the
+  number, and vary sentence length. If the first line could open any PR on any repo,
+  delete it.
+- Treat `AGENTS.md`, `CLAUDE.md`, `GEMINI.md`, and `.github/copilot-instructions.md`
+  as review-required: they may be intentional team config, so confirm before
+  committing them.
+
+Project map:
+
+- `src/scan.mjs` and `src/rules.mjs` load rules and make policy decisions.
+- `src/hook.mjs` and `src/githooks.mjs` enforce them in agent and Git hooks.
+- `src/state.mjs` owns local policy state. `bin/aimhooman.mjs` is the CLI.
+- `tests/` mirrors the production paths and includes real temporary Git repos.
+
+Use Node.js 22.8 or newer. Before finishing a change, run:
+
+```sh
+npm run check
+npm test
+npm run test:coverage
+```
+
+aimhooman also enforces this at commit time, so a blocked commit means a real
+violation to fix, not a check to bypass. The rule is simple: AI works, hoomans
+ship.
+<!-- aimhooman:ruleset-end -->

package/CHANGELOG.md

@@ -0,0 +1,153 @@
+# Changelog
+
+All notable changes to this project are documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [Unreleased]
+
+## [0.1.0] - 2026-07-15
+
+First public release. aimhooman keeps AI coding-agent residue out of Git history
+without getting in your way: AI session/state files, secrets, AI attribution in
+commit messages, and AI markers left in code.
+
+### Fixed
+
+- Scan staged blob content when an agent tries to bypass `pre-commit` under the
+  clean or compliance profile.
+- Treat earlier repository or hook mutations as a possible guard bypass, track
+  index changes such as `git mv`, and preserve policy-transition risk through
+  Git aliases before a compound commit.
+- Treat branch-writing `fetch`, `worktree`, `stash`, `bisect`, `remote`, and local
+  `push` receiver flows as protected ref mutations; reject explicit or configured
+  receive-pack indirection before it can bypass the final hook.
+- Require the managed final guards to be present for protected agent Git
+  operations on every profile, not only `strict`.
+- Ignore local Git replacement refs in policy, history, authorization, and audit
+  reads; treat `git replace` as a policy transition in compound commands.
+- Scan every index stage during unresolved conflicts so a secret on either side
+  cannot be missed.
+- Authorize protected-path CI only when the exact workflow-run attempt's actor and
+  triggering actor, including their IDs, match the current personal-repository
+  owner returned by GitHub. Bind that authorization to exact commits, paths,
+  blobs, modes, deletion tombstones, and policy migrations; non-owner changes fail
+  closed with no reviewer fallback.
+- Keep the published local rule-pack schema compatible with strict JSON Schema
+  validators.
+- Prove npm version absence and forward dist-tag movement with bounded retries
+  before publish; retry integrity, dist-tag, and provenance reads afterward,
+  including idempotent reruns.
+- Classify SemVer build metadata correctly when creating GitHub releases.
+- Run GitHub Copilot hooks on PowerShell as well as Bash without swallowing
+  aimhooman's exit status.
+- Render native Windows `PATH` entries in Git Bash syntax inside managed hooks,
+  and read reference updates with shell builtins so the hooks do not depend on
+  semicolon path parsing or external core utilities.
+- Reject shell-expanded and split-worktree targets before alias or policy lookup.
+  Leading-tilde forms fail closed on every platform, exactly-two-slash targets
+  fail closed on POSIX, and guarded Windows changes reject POSIX-root,
+  drive-relative, and incomplete UNC targets.
+- Treat non-POSIX executor payloads as uncertain and reject their repository-
+  selection syntax before a protected operation can inspect the wrong policy.
+- Preserve outer runtime and Git-target environment risks inside nested shell
+  payloads, including Bash, POSIX-shell, and Zsh startup-file variables.
+- Reject guarded Git operations wrapped by commands that can select another cwd
+  or filesystem namespace, such as `sudo`, `chroot`, WSL, and sandbox launchers.
+- Reject nested non-POSIX shells and explicit login, interactive, or startup-file
+  shell launches before they can redirect repository policy lookup.
+- Keep concurrent lifecycle commands queued long enough for the earlier command
+  to finish, including the case where that command fails and its successor retries.
+- Leave unsupported bare repositories unchanged when they inherit the optional
+  global hook directory.
+- Accept Git 2.54's early `preparing` reference-transaction callback while
+  checking guard integrity there and keeping the full scan in the locked
+  `prepared` phase.
+- Provision pinned Go 1.25 in test and release workflows so actionlint runs on
+  macOS runner images that do not preinstall Go.
+- Build the package-manifest private-key fixture at test runtime so a fresh
+  strict history scan does not mistake the test source for credential material.
+- Scan from empty history when a force-push event names an old commit that is no
+  longer fetchable, covering every commit reachable from the replacement head.
+- Allow public PEM certificates in the package manifest while rejecting PEM
+  files that contain a private-key header.
+- Publish to npm from a `v*` tag push: the release workflow installs dependencies,
+  runs the test suite, and publishes with npm build provenance authenticated by the
+  `NPM_TOKEN` secret. Use Node 24 and SHA-pinned actions.
+
+### Overview
+
+- Vendor-neutral guard with one detection core and a thin adapter per host.
+- Repair-first ordinary commits by default: session-start auto-exclude writes known AI
+  artifacts to `.git/info/exclude`; `pre-commit` unstages any that slip through;
+  `commit-msg` removes complete exact high-confidence attribution lines and reports
+  broader candidates without editing them. An unterminated exact final line stops
+  unchanged because a byte-safe repair cannot be proved.
+- 29 built-in rules across paths, attribution, markers, and secrets.
+- Three policy profiles: `clean` (default — repair-first on the ordinary commit path),
+  `strict` (opt-in hard enforcement that cancels the commit), and `compliance`
+  (repair-first but keeps any AI disclosure a regulated policy requires).
+- Any block that cannot be unstaged or repaired stops the commit; a pinned-tree
+  and final full-snapshot scan prevent a pre-existing tracked block from riding
+  through an unrelated change.
+
+### Enforcement surfaces
+
+- SessionStart auto-exclude for known AI artifacts (local, never `.gitignore`).
+- `pre-commit` and `pre-merge-commit` unstage of AI session/state files and
+  secrets from the index.
+- `commit-msg` removal of complete exact AI attribution lines.
+- Prepared `reference-transaction` full-scan of every commit introduced by an
+  ordinary commit, cherry-pick, revert, rebase, `git am`, fetch/worktree branch
+  creation, or direct ref update.
+- `PreToolUse` path advisory plus fail-closed checks on every profile for empty,
+  invalid, or non-object hook JSON, missing managed guards, unresolved Git
+  commands/aliases, and hook or receive-pack indirection around protected ref
+  mutations. Unknown executor argument shapes fail closed on `strict`, which also
+  rejects uncertain commit execution and `git commit --no-verify`.
+- CLI targets: `--staged`, `--tracked`, `--commit <rev>`, `--range <base>...<head>`,
+  `--message <file>`, plus `audit` and its `scan` alias as shorthand for a full
+  tracked-index scan.
+- Optional `aimhooman init --global --yes` guards terminal Git in eligible non-bare
+  repositories that do not override `core.hooksPath` locally.
+
+### Host adapters
+
+- Plugin-tier (live `PreToolUse` guard): Claude Code, Codex, GitHub Copilot CLI.
+- Instruction-tier (ruleset text): Cursor, Cline, Windsurf, Kiro, Gemini CLI,
+  Gemini Code Assist, Google Antigravity, plus `AGENTS.md` as a universal fallback.
+- Host registry in `docs/hosts.json` records adapter, activation contract, and
+  verification status for each host.
+
+### Configuration
+
+- Versioned `.aimhooman.json` project profiles with precedence over per-clone
+  defaults; an individual `check` may escalate but cannot weaken the team profile.
+- Per-clone local rule packs in `<git-common-dir>/aimhooman/rules/*.json`
+  (core-first; local rules only add detection, never weaken a built-in block).
+- Per-decision overrides (`allow` / `deny`) in
+  `<git-common-dir>/aimhooman/overrides.json` (local, never committed).
+- Malformed project policy fails closed with an actionable error.
+
+### Engineering
+
+- Zero-runtime-dependency Node ESM (Node 22.8+ and Git 2.28+), shipped as source.
+- State lives in the git dir. Init normally adds no worktree file; when a
+  repository already uses a worktree-relative `core.hooksPath`, that configured
+  directory remains the hook location. Foreign Git hooks are chained and
+  restored on uninstall.
+- Published JSON schemas; machine reports include target policy identity,
+  completeness, scan statistics, and commit/object metadata.
+- Release pipeline publishes to npm on a `v*` tag push with SHA-pinned actions and
+  npm build provenance.
+
+### Security
+
+- Malformed local rule packs are skipped with a warning on `clean`; `strict` fails
+  closed. Corrupt enforcement state and unreadable Git targets return an error.
+- `init --global` refuses to clobber a pre-existing `core.hooksPath`.
+- The `PreToolUse` guard intercepts `git commit --no-verify` upstream of git, which
+  a git hook cannot.
+- Hardened command parser (quoted verbs, newline separators, pathed git binaries),
+  bounded path-glob matching with character classes, and hook input validation.