codehere 2.0.0-beta.2 → 2.0.0

package/README.md

@@ -2,7 +2,7 @@
 
 **A team brain for vibe coders.** It stops your AI coding agent from going in circles, checks "done" for real, and shares what your team learns. It works through plain files in git.
 
-Codehere is built to remove work, not add it. You keep using Claude Code or Codex CLI as usual. Codehere talks to the agent, not to you, and stays silent unless it helps.
+Codehere is built to remove work, not add it. You keep using Claude Code, Codex CLI or Copilot CLI as usual. Codehere talks to the agent, not to you, and stays silent unless it helps.
 
 ## What it does
 
@@ -21,19 +21,21 @@ Codehere is built to remove work, not add it. You keep using Claude Code or Code
 ## Install (2 minutes)
 
 ```bash
-npm install -g codehere@next   # the 2.0 beta; codehere@latest is still 1.x
+npm install -g codehere
 cd your-project
-codehere init --agent claude,codex   # finds your test command, adds the hooks, asks you to approve the command
-git add .codehere .claude/settings.json .codex/hooks.json .gitignore
+codehere init --agent claude,codex,copilot   # finds your test command, adds the hooks, asks you to approve the command
+git add .codehere .claude/settings.json .codex/hooks.json .github/hooks/codehere.json .gitignore
 git commit -m "Add Codehere" && git push
 ```
 
-Use `--agent claude` or `--agent codex` for one agent; with no `--agent`, `init` sets up Claude Code. The hooks go in `.claude/settings.json` and `.codex/hooks.json`.
+Name only the agents your team uses, such as `--agent claude` or `--agent codex,copilot`; with no `--agent`, `init` sets up Claude Code. The hooks go in `.claude/settings.json`, `.codex/hooks.json` and `.github/hooks/codehere.json`.
 
-Each teammate installs the CLI once (`npm install -g codehere@next`) and runs `codehere approve` once in the repo. Until they do, their agent tells them at session start that the test check is off. The hooks and the brain arrive through git.
+Each teammate installs the CLI once (`npm install -g codehere`) and runs `codehere approve` once in the repo. Until they do, their agent tells them at session start that the test check is off. The hooks and the brain arrive through git.
 
 Codex runs a repo's hooks only after each developer trusts them. Open Codex in the repo, trust the project if asked, then type `/hooks` and trust the four Codehere hooks. Do it once per clone, and again after an update that changes the hooks. Until then Codex skips them, and `codex exec` does not warn.
 
+Copilot runs a repo's hooks only in a folder you trust: answer yes when Copilot asks on its first start in the repo. Copilot also runs the hooks in `.claude/settings.json`. Where `.github/hooks/codehere.json` runs Codehere, those stay quiet, so nothing runs twice; without it, they give Copilot the same checks.
+
 `codehere uninstall` removes the hooks, and `.codehere/` too when it holds no rules, lessons, help cards or briefs.
 
 The test command is in `.codehere/config.json`, which anyone with push access can change. So Codehere runs it only after you approve that exact command on your machine, at your terminal. If a pull or the agent changes it, the finish check skips it and tells you, until you approve it again. Approvals are kept in `~/.config/codehere/approvals.json`, outside the repo, so nothing that arrives through git can approve a command.
@@ -69,25 +71,25 @@ Don't want to commit help cards? Copy the printed text into WhatsApp, Discord or
 
 ## Which agents
 
-| | Claude Code | Codex CLI |
-|---|---|---|
-| Team rules at session start | yes | yes |
-| A teammate's lesson after a failed command | yes | yes |
-| "Same error 3 times: stop and find the root cause" | yes | yes, given next to the error output |
-| Context added to a short prompt such as "still broken" | yes | yes |
-| Finish check: your tests re-run, the agent sent back | yes | yes |
-| Your one line (✓ ✕ ⊘ ›) | yes | in the interactive screen, as `↳ Hook · codehere ✓ …`; not in `codex exec` |
-| Restart-brief suggestion in long sessions | yes | no |
-| Before it works | nothing | each developer trusts the hooks in `/hooks` |
-| MCP tools, instruction file | no | no |
-| Tested on | 2.1.311 with haiku: [result](docs/conformance/2026-10-27-claude-2.1.311-haiku.md) | 0.164.0 with a local model (Qwen 3.5 9B through Ollama): [result](docs/conformance/2026-10-27-codex-0.164.0-qwen3.5-9b-q4-K-M.md) |
+| | Claude Code | Codex CLI | Copilot CLI |
+|---|---|---|---|
+| Team rules at session start | yes | yes | yes |
+| A teammate's lesson after a failed command | yes | yes | yes |
+| "Same error 3 times: stop and find the root cause" | yes | yes, given next to the error output | yes, given next to the error output |
+| Context added to a short prompt such as "still broken" | yes | yes | no: by Copilot's docs it drops a prompt hook's output |
+| Finish check: your tests re-run, the agent sent back | yes | yes | yes |
+| Your one line (✓ ✕ ⊘ ›) | yes | in the interactive screen, as `↳ Hook · codehere ✓ …`; not in `codex exec` | no; when the agent is sent back, you see the full reason as the next prompt |
+| Restart-brief suggestion in long sessions | yes | no | no |
+| Before it works | nothing | each developer trusts the hooks in `/hooks` | each developer trusts the folder when Copilot asks |
+| MCP tools, instruction file | no | no | no |
+| Tested on | 2.1.311 with haiku: [result](docs/conformance/2026-10-27-claude-2.1.311-haiku.md) | 0.164.0 with a local model (Qwen 3.5 9B through Ollama): [result](docs/conformance/2026-10-27-codex-0.164.0-qwen3.5-9b-q4-K-M.md) | 1.0.105 with a local model (Qwen 3.5 9B through Ollama): [result](docs/conformance/2026-10-29-copilot-1.0.105-qwen3.5-9b-q4-K-M.md), with only the Claude Code hooks installed: [result](docs/conformance/2026-10-29-copilot-1.0.105-qwen3.5-9b-q4-K-M-claude-hooks-only.md), and with a GitHub-hosted model (`auto`, which picked gpt-6-luna): [result](docs/conformance/2026-10-29-copilot-1.0.105-auto.md) |
 
 The rows for rules, lessons and the finish check are checked on the real agent by `node scripts/conformance.mjs`. The other rows are checked by `npm test` on hook inputs recorded from each agent.
 
 ## Commands
 
 ```
-codehere init [--agent claude,codex] [--test "npm test"]
+codehere init [--agent claude,codex,copilot] [--test "npm test"]
                                         set up (safe to run again)
 codehere approve                        let the finish check run the test command on this machine
 codehere uninstall                      remove Codehere from this repo (asks first)
@@ -135,17 +137,19 @@ node demo/demo.mjs --step   # press Enter between scenes (good for a live talk)
 npm test                    # end-to-end checks
 ```
 
-## Known limits (2.0 beta)
+## Known limits
 
-- Claude Code and Codex CLI. Copilot and OpenCode are next. See [Which agents](#which-agents) for what works on each.
+- Claude Code, Codex CLI and Copilot CLI. OpenCode is next. Copilot in VS Code reads the same hook files but has not been checked, and Copilot's cloud agent is not supported. See [Which agents](#which-agents) for what works on each.
 - Lessons match by error text. Very generic errors can bring up a lesson that doesn't fit, which is why the agent gets it as reference, not as an instruction.
 - Redaction catches common keys, tokens and emails. Always read a help card before sharing it.
 - The "done" check needs a test command. Without one, only the tampering check runs.
 - Approvals are per folder. Each git worktree or extra clone needs its own `codehere approve`.
 - In `codex exec`, Codex shows you none of Codehere's lines. The agent still gets every message, and the finish check still sends it back.
-- `codehere approve` and `codehere uninstall` ask only in your own terminal. They refuse when Claude Code or Codex is among the processes above them, and inside Codex's sandbox. An agent that detaches its terminal on purpose can still answer, so the finish check of a session that was open tells you when an approval or an uninstall happened.
-- Approving `npm test` approves that command, not the scripts behind it in `package.json`. And Claude Code runs any hook a teammate adds to the committed `.claude/settings.json`; Codex asks each developer to trust a new or changed hook first. Both are your repo's code, with the same trust as running it yourself.
-- When the agent changes `.codehere/config.json`, `.codehere/rules.md`, the hook and env settings in `.claude/settings*.json`, or `.codex/hooks.json` or `.codex/config.toml` during a session, the finish check tells you. It cannot tell you if the session ends without a finish check, for example when you close it mid-turn. An agent that sets out to fake your terminal or rewrite Codehere's local files can also get around this.
+- Copilot shows you none of Codehere's lines either, so on Copilot you do not see the ✓, the ⊘, or the notes about an approval, an uninstall or a settings change during a session. The agent still gets every message, and when the finish check sends it back you see the reason as the next prompt.
+- Copilot's cloud agent also runs the committed `.github/hooks/codehere.json`. `codehere` is not installed there, so by GitHub's docs each hook fails and the job carries on without Codehere.
+- `codehere approve` and `codehere uninstall` ask only in your own terminal. They refuse when Claude Code, Codex or Copilot is among the processes above them, and inside Codex's sandbox. An agent that detaches its terminal on purpose can still answer, so the finish check of a session that was open tells you when an approval or an uninstall happened (not on Copilot, which shows none of these lines).
+- Approving `npm test` approves that command, not the scripts behind it in `package.json`. And Claude Code runs any hook a teammate adds to the committed `.claude/settings.json`, as Copilot does in a trusted folder for `.github/hooks/` and `.claude/settings.json`; Codex asks each developer to trust a new or changed hook first. Both are your repo's code, with the same trust as running it yourself.
+- When the agent changes `.codehere/config.json`, `.codehere/rules.md`, the hook and env settings in `.claude/settings*.json` or `.github/copilot/settings*.json`, or `.codex/hooks.json`, `.codex/config.toml` or `.github/hooks/codehere.json` during a session, the finish check tells you. It cannot tell you if the session ends without a finish check, for example when you close it mid-turn. An agent that sets out to fake your terminal or rewrite Codehere's local files can also get around this.
 - Only the test command needs approval. A teammate who sets `verifyOnStop` to `false` in `.codehere/config.json` turns the finish check off for everyone; `codehere status` shows it as off.
 - Sharing is through git. There is no server, so teammates see updates after `git pull`.
 - Prompt checks are plain word lists (English and Indonesian). They can miss a short prompt in another language.

package/bin/codehere.js

@@ -25,7 +25,7 @@ const [command, ...rest] = pos;
 const USAGE = `codehere · a git-native team brain for vibe coders
 
 Setup
-  codehere init [--agent claude,codex] [--test "npm test"]
+  codehere init [--agent claude,codex,copilot] [--test "npm test"]
                                       set up .codehere/ and the agents' hooks (default: Claude Code)
   codehere approve                    let the finish check run the test command on this machine
   codehere uninstall                  remove Codehere's hooks, and .codehere/ if it holds no team content
@@ -48,7 +48,7 @@ When you're stuck
   codehere stats                      turns to green, loops stopped, session size
 
 Hooks (called by the agent, not by you)
-  codehere hook session-start | prompt | post-tool | stop [--agent codex]
+  codehere hook session-start | prompt | post-tool | stop [--agent codex|copilot]
 `;
 
 const run = async () => {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "codehere",
-  "version": "2.0.0-beta.2",
+  "version": "2.0.0",
   "description": "A git-native team brain for vibe coders. Silent fixes for AI coding agents: breaks doom loops, verifies 'done', blocks test tampering, and shares what your team learned.",
   "repository": {
     "type": "git",

package/src/commands.js

@@ -28,12 +28,24 @@ const CODEX_HOOK_EVENTS = {
   Stop: { cmd: 'codehere hook stop --agent codex', timeout: 180 },
 };
 
+// Copilot's own format: camelCase events, hooks listed directly under each event, timeout as
+// timeoutSec. Its prompt output is dropped, so that hook only counts the turn.
+const COPILOT_HOOK_EVENTS = {
+  sessionStart: { cmd: 'codehere hook session-start --agent copilot' },
+  userPromptSubmitted: { cmd: 'codehere hook prompt --agent copilot', timeout: 10 },
+  postToolUse: { cmd: 'codehere hook post-tool --agent copilot', matcher: 'bash' },
+  agentStop: { cmd: 'codehere hook stop --agent copilot', timeout: 180 },
+};
+
 const AGENTS = {
   claude: { name: 'Claude Code', file: '.claude/settings.json', events: HOOK_EVENTS },
   codex: { name: 'Codex CLI', file: '.codex/hooks.json', events: CODEX_HOOK_EVENTS },
+  copilot: { name: 'Copilot CLI', file: '.github/hooks/codehere.json', events: COPILOT_HOOK_EVENTS, flat: true },
 };
 
 const ours = (h) => typeof h?.command === 'string' && h.command.startsWith('codehere hook ');
+// Claude Code and Codex wrap hooks in groups; Copilot lists each hook directly.
+const hooksOf = (g) => (Array.isArray(g?.hooks) ? g.hooks : [g]);
 
 function requireRoot() {
   const root = findRoot();
@@ -57,7 +69,6 @@ function detectTestCommand(root) {
   return null;
 }
 
-// Claude Code's .claude/settings.json and Codex's .codex/hooks.json share this shape.
 function mergeHooks(root, agent) {
   const file = path.join(root, agent.file);
   fs.mkdirSync(path.dirname(file), { recursive: true });
@@ -65,12 +76,19 @@ function mergeHooks(root, agent) {
   try {
     settings = JSON.parse(fs.readFileSync(file, 'utf8'));
   } catch {}
+  if (agent.flat) settings = { version: 1, ...settings };
   settings.hooks = settings.hooks || {};
   for (const [event, spec] of Object.entries(agent.events)) {
     const list = (settings.hooks[event] = settings.hooks[event] || []);
-    const already = list.some((g) => (g?.hooks || []).some((h) => h?.command === spec.cmd));
+    const already = list.some((g) => hooksOf(g).some((h) => h?.command === spec.cmd));
     if (already) continue;
     const hook = { type: 'command', command: spec.cmd };
+    if (agent.flat) {
+      if (spec.matcher) hook.matcher = spec.matcher;
+      if (spec.timeout) hook.timeoutSec = spec.timeout;
+      list.push(hook);
+      continue;
+    }
     if (spec.timeout) hook.timeout = spec.timeout;
     const group = { hooks: [hook] };
     if (spec.matcher) group.matcher = spec.matcher;
@@ -80,7 +98,7 @@ function mergeHooks(root, agent) {
 }
 
 // Takes Codehere's hooks out of the agent's settings file, keeping everything else.
-// A file left with nothing in it is removed, and its folder if that is then empty.
+// A file left with nothing in it is removed, and its folders up to the repo if they are then empty.
 function removeHooks(root, agent) {
   const file = path.join(root, agent.file);
   let settings;
@@ -95,24 +113,27 @@ function removeHooks(root, agent) {
     if (!Array.isArray(hooks[event])) continue;
     hooks[event] = hooks[event]
       .filter((g) => g && typeof g === 'object')
-      .map((g) => {
-        const all = Array.isArray(g.hooks) ? g.hooks : [];
-        const kept = all.filter((h) => !ours(h));
-        if (kept.length !== all.length) found = true;
-        return { ...g, hooks: kept };
-      })
-      .filter((g) => g.hooks.length);
+      .flatMap((g) => {
+        if (!Array.isArray(g.hooks)) return ours(g) ? ((found = true), []) : [g];
+        const kept = g.hooks.filter((h) => !ours(h));
+        if (kept.length !== g.hooks.length) found = true;
+        return kept.length ? [{ ...g, hooks: kept }] : [];
+      });
     if (!hooks[event].length) delete hooks[event];
   }
   if (!found) return false;
   if (!Object.keys(hooks).length) delete settings.hooks;
-  if (Object.keys(settings).length) {
+  if (Object.keys(settings).some((k) => !(agent.flat && k === 'version'))) {
     fs.writeFileSync(file, JSON.stringify(settings, null, 2) + '\n');
   } else {
     fs.rmSync(file);
-    try {
-      fs.rmdirSync(path.dirname(file));
-    } catch {} // not empty: something else lives there
+    for (let d = path.dirname(file); d !== root; d = path.dirname(d)) {
+      try {
+        fs.rmdirSync(d);
+      } catch {
+        break; // not empty: something else lives there
+      }
+    }
   }
   return true;
 }
@@ -121,7 +142,7 @@ function installedAgents(root) {
   return Object.values(AGENTS).filter((a) => {
     try {
       const hooks = JSON.parse(fs.readFileSync(path.join(root, a.file), 'utf8')).hooks || {};
-      return Object.values(hooks).some((gs) => Array.isArray(gs) && gs.some((g) => Array.isArray(g?.hooks) && g.hooks.some(ours)));
+      return Object.values(hooks).some((gs) => Array.isArray(gs) && gs.some((g) => hooksOf(g).some(ours)));
     } catch {
       return false;
     }
@@ -149,14 +170,16 @@ const PS = fs.existsSync('/bin/ps') ? '/bin/ps' : '/usr/bin/ps';
 function underAgent() {
   let pid = process.ppid;
   for (let i = 0; i < 100 && pid > 1; i++) {
-    let line;
+    let line, args;
     try {
       line = execFileSync(PS, ['-o', 'ppid=,comm=', '-p', String(pid)], { encoding: 'utf8', stdio: ['ignore', 'pipe', 'ignore'] }).trim();
+      args = execFileSync(PS, ['-o', 'args=', '-p', String(pid)], { encoding: 'utf8', stdio: ['ignore', 'pipe', 'ignore'] }).trim();
     } catch {
       return true;
     }
     const m = /^(\d+)\s+(.+)$/.exec(line);
-    if (!m || AGENT_PROCESSES.has(path.basename(m[2]).toLowerCase())) return true;
+    // on Linux, comm can be a thread name and not the program (seen with Node 24), so argv[0] counts too
+    if (!m || [m[2], args.split(/\s/, 1)[0]].some((n) => AGENT_PROCESSES.has(path.basename(n).toLowerCase()))) return true;
     pid = Number(m[1]);
   }
   return pid > 1;
@@ -217,7 +240,7 @@ const README_MD = [
 export async function init(flags) {
   const names = flags.agent === undefined ? ['claude'] : String(flags.agent === true ? '' : flags.agent).split(',').map((x) => x.trim().toLowerCase());
   if (!names.every((n) => Object.hasOwn(AGENTS, n))) {
-    console.error(`Unknown agent "${printable(flags.agent === true ? '' : flags.agent)}". Use --agent claude, --agent codex or --agent claude,codex.`);
+    console.error(`Unknown agent "${printable(flags.agent === true ? '' : flags.agent)}". Use --agent with claude, codex or copilot, or several: --agent claude,codex,copilot.`);
     process.exit(1);
   }
   const agents = [...new Set(names)].map((n) => AGENTS[n]);
@@ -253,6 +276,10 @@ export async function init(flags) {
     console.log('then type /hooks and trust the four Codehere hooks. Until then Codex skips them, and `codex exec` does not warn.');
     console.log('');
   }
+  if (agents.includes(AGENTS.copilot)) {
+    console.log("Copilot runs a repo's hooks only in a folder you trust. Answer yes when Copilot asks on its first start here.");
+    console.log('');
+  }
   console.log(`Next: commit .codehere/ and ${agents.map((a) => a.file).join(' and ')} so your team gets it too.`);
   console.log(`Then just use ${agents.map((a) => a.name).join(' or ')} as usual. Codehere stays quiet unless it helps.`);
 }

package/src/hooks.js

@@ -1,5 +1,6 @@
-// Hook handlers for Claude Code and Codex CLI (--agent codex). Every handler fails open: if
-// anything goes wrong, it exits 0 and the agent carries on as if Codehere were not installed.
+// Hook handlers for Claude Code, Codex CLI (--agent codex) and Copilot CLI (--agent copilot).
+// Every handler fails open: if anything goes wrong, it exits 0 and the agent carries on as if
+// Codehere were not installed.
 import fs from 'node:fs';
 import path from 'node:path';
 import crypto from 'node:crypto';
@@ -35,10 +36,44 @@ function toAgent(message) {
 
 const quiet = () => process.exit(0);
 
+// Copilot's inputs: camelCase from its own hook file, snake_case from .claude/settings.json.
+// Map them to the names the handlers read. A failed command arrives as a successful tool
+// call; only its output shows the failure.
+function fromCopilot(i) {
+  const tool = i.toolName ?? i.tool_name;
+  return {
+    ...i,
+    session_id: i.sessionId ?? i.session_id,
+    tool_name: tool === 'bash' ? 'Bash' : tool,
+    tool_input: i.toolArgs ?? i.tool_input,
+    tool_response: i.toolResult?.textResultForLlm ?? i.tool_result?.text_result_for_llm,
+  };
+}
+
+const COPILOT_EVENT = { 'session-start': 'sessionStart', prompt: 'userPromptSubmitted', 'post-tool': 'postToolUse', stop: 'agentStop' };
+
+// Whether Copilot's own hook file runs Codehere for this event.
+function copilotFileRuns(root, event) {
+  try {
+    const hooks = JSON.parse(fs.readFileSync(path.join(root, '.github/hooks/codehere.json'), 'utf8')).hooks;
+    return (hooks?.[COPILOT_EVENT[event]] || []).some((h) => h?.command === `codehere hook ${event} --agent copilot`);
+  } catch {
+    return false;
+  }
+}
+
 export async function runHook(event, agent) {
-  const input = readStdin();
+  let input = readStdin();
   const start = input.cwd || process.env.CLAUDE_PROJECT_DIR || process.cwd();
   const root = findRoot(start);
+  // Copilot also runs the hooks in .claude/settings.json, adding a `timestamp` Claude Code does
+  // not send. When Copilot's own file runs Codehere for this event, these stay quiet so nothing
+  // runs twice. Otherwise they are all Copilot has, and answer in its format.
+  if (agent !== 'copilot' && process.env.COPILOT_CLI && 'timestamp' in input) {
+    if (root && copilotFileRuns(root, event)) quiet();
+    agent = 'copilot';
+  }
+  if (agent === 'copilot') input = fromCopilot(input);
   if (!root) {
     // uninstall ran while this session was open: its hooks still run, .codehere/ is gone
     const at = event === 'stop' && removedAt(start);
@@ -46,9 +81,9 @@ export async function runHook(event, agent) {
     quiet();
   }
   try {
-    if (event === 'session-start') return sessionStart(root, input);
+    if (event === 'session-start') return sessionStart(root, input, agent);
     if (event === 'prompt') return promptSubmit(root, input);
-    if (event === 'post-tool') return postTool(root, input, agent === 'codex' ? 'codex' : 'claude');
+    if (event === 'post-tool') return postTool(root, input, agent === 'codex' || agent === 'copilot' ? agent : 'claude');
     if (event === 'stop') return stop(root, input);
   } catch (err) {
     if (process.env.CODEHERE_DEBUG) console.error(err);
@@ -58,7 +93,7 @@ export async function runHook(event, agent) {
 
 // ---------- SessionStart: load only human-confirmed context ----------
 
-function sessionStart(root, input) {
+function sessionStart(root, input, agent) {
   const state = readState(root);
   const s = session(state, input.session_id);
   s.transcript = input.transcript_path || s.transcript;
@@ -110,6 +145,8 @@ function sessionStart(root, input) {
   }
 
   writeState(root, state);
+  // Copilot reads only JSON here; Claude Code and Codex take the plain text
+  if (out.length && agent === 'copilot') emitJson({ additionalContext: out.join('\n') });
   if (out.length) process.stdout.write(out.join('\n') + '\n');
   quiet();
 }
@@ -125,7 +162,7 @@ function briefFile(root, file) {
 
 // ---------- what changed in THIS session ----------
 
-const IGNORE = (f) => f.startsWith('.codehere/') || f.startsWith('.claude/') || f.startsWith('.codex/');
+const IGNORE = (f) => ['.codehere/', '.claude/', '.codex/', '.github/copilot/', '.github/hooks/'].some((d) => f.startsWith(d));
 
 export function dirtySnapshot(root) {
   const files = statusFiles(root).filter((f) => !IGNORE(f));
@@ -155,21 +192,33 @@ function changedThisSession(root, s) {
 
 // Files that decide what Codehere runs and tells every agent, and the agents' own hook
 // settings. Claude Code and Codex load hooks at session start, so a hook removed mid-session
-// still runs at this session's finish check and can report it.
-const PROTECTED = ['.codehere/config.json', '.codehere/rules.md', '.claude/settings.json', '.claude/settings.local.json', '.codex/hooks.json', '.codex/config.toml'];
+// still runs at this session's finish check and can report it. Copilot runs every file in
+// .github/hooks/, so that folder counts as one.
+const PROTECTED = [
+  '.codehere/config.json', '.codehere/rules.md', '.claude/settings.json', '.claude/settings.local.json', '.codex/hooks.json', '.codex/config.toml',
+  '.github/hooks/', '.github/copilot/settings.json', '.github/copilot/settings.local.json',
+];
+
+function readProtected(root, f) {
+  if (!f.endsWith('/')) return fs.readFileSync(path.join(root, f), 'utf8');
+  const files = fs.readdirSync(path.join(root, f)).filter((n) => n.endsWith('.json')).sort();
+  if (!files.length) return null;
+  return files.map((n) => `${n}\0${fs.readFileSync(path.join(root, f, n), 'utf8')}`).join('\0');
+}
 
 function settingsSnapshot(root) {
   const snap = {};
   for (const f of PROTECTED) {
     let text = null;
     try {
-      text = fs.readFileSync(path.join(root, f), 'utf8');
+      text = readProtected(root, f);
     } catch {}
     // In .claude/ only the hook and env settings count: Claude Code also saves permission
     // rules there when the developer picks "don't ask again". env can move HOME or
     // XDG_CONFIG_HOME, where approvals are kept. A missing file has none of them.
     // Codex keeps what it saves in the user's own config, so .codex/ files count whole.
-    if (f.startsWith('.claude/')) {
+    // Copilot's settings files in .github/copilot/ count like .claude/; its hooks file is ours.
+    if (f.startsWith('.claude/') || f.startsWith('.github/copilot/')) {
       try {
         const j = JSON.parse(text ?? '{}');
         text = JSON.stringify([j.hooks ?? null, j.disableAllHooks ?? null, j.env ?? null]);
@@ -351,7 +400,9 @@ function postTool(root, input, agent) {
       );
     }
     // Codex replaces the command's output with a block's text (or exit 2's), so the agent
-    // would lose the error. There it goes in as context next to the output.
+    // would lose the error. There it goes in as context next to the output. Copilot shows
+    // exit 2 to the developer, not the agent, and takes context only in its own format.
+    if (agent === 'copilot') emitJson({ additionalContext: msg.join('\n') });
     if (agent === 'codex') {
       emitJson({
         ...(lesson ? { systemMessage: lessonLine(lesson) } : {}),
@@ -363,6 +414,7 @@ function postTool(root, input, agent) {
     toAgent(msg.join('\n') + '\n');
   }
 
+  if (lesson && agent === 'copilot') emitJson({ additionalContext: lessonText(root, lesson) });
   if (lesson) {
     emitJson({
       systemMessage: lessonLine(lesson),