AI Meeting Skill

Example 1: Quick Product Decision Review

cat > /tmp/ai-meeting-brief.md <<'EOF' # Decision Should we add team workspaces to the app in the next release? # Context - Users currently share one personal workspace. - Enterprise prospects ask for shared projects and role-based access. - The team has two weeks before feature freeze. # Evaluation Criteria - Real user value - Implementation cost - Security and permission risk - Whether a smaller MVP exists EOF MEETING_DIR=$(node ai-meeting/scripts/ai-meeting.mjs create \ --topic "Team workspaces release decision" \ --brief-file /tmp/ai-meeting-brief.md \ --agents builder:codex,critic:claude,architect:codex \ | tee /tmp/ai-meeting-create.json \ | node -e 'let s=""; process.stdin.on("data", d => s += d); process.stdin.on("end", () => console.log(JSON.parse(s).meetingDir));') node ai-meeting/scripts/ai-meeting.mjs round --meeting-dir "$MEETING_DIR" --round 1 node ai-meeting/scripts/ai-meeting.mjs round --meeting-dir "$MEETING_DIR" --round 2 node ai-meeting/scripts/ai-meeting.mjs synthesize --meeting-dir "$MEETING_DIR"

sed -n '1,220p' "$MEETING_DIR/synthesis/final.md"

Example 2: Review A Development Document With Materials

cat > /tmp/ai-meeting-brief.md <<'EOF' # Decision Review the proposed API redesign and decide whether it is ready for implementation. # What To Focus On - Developer experience - Backward compatibility - Migration risk - Test coverage - Simpler alternatives EOF mkdir -p /tmp/ai-meeting-materials cat > /tmp/ai-meeting-materials/api-redesign.md <<'EOF' # API Redesign Replace the old /v1/tasks response envelope with a flatter /v2/tasks shape. Support both versions for one release. Add migration warnings to SDKs. EOF cat > /tmp/ai-meeting-materials/migration-plan.md <<'EOF' # Migration Plan 1. Ship /v2/tasks behind a beta flag. 2. Add SDK compatibility helpers. 3. Deprecate /v1/tasks after telemetry shows 90% migration. EOF MEETING_DIR=$(node ai-meeting/scripts/ai-meeting.mjs create \ --topic "API redesign readiness review" \ --brief-file /tmp/ai-meeting-brief.md \ --material README.md \ --material /tmp/ai-meeting-materials/api-redesign.md \ --material /tmp/ai-meeting-materials/migration-plan.md \ --agents builder:codex,critic:claude,user-advocate:codex,architect:claude \ | tee /tmp/ai-meeting-create.json \ | node -e 'let s=""; process.stdin.on("data", d => s += d); process.stdin.on("end", () => console.log(JSON.parse(s).meetingDir));') node ai-meeting/scripts/ai-meeting.mjs round --meeting-dir "$MEETING_DIR" --round 1 node ai-meeting/scripts/ai-meeting.mjs round --meeting-dir "$MEETING_DIR" --round 2 node ai-meeting/scripts/ai-meeting.mjs synthesize --meeting-dir "$MEETING_DIR"

Materials are copied into the meeting directory and injected into prompts as untrusted data blocks. Large materials are marked withtruncatedForPrompt=true; the final report must treat that as an evidence gap.

Example 3: Preview Prompts Before Spending Tokens

cat > /tmp/ai-meeting-brief.md <<'EOF' # Decision Should we refactor the background job system now or defer it? # Criteria - Reliability improvement - Risk of regressions - Amount of code churn - Operational impact EOF MEETING_DIR=$(node ai-meeting/scripts/ai-meeting.mjs create \ --topic "Background job refactor decision" \ --brief-file /tmp/ai-meeting-brief.md \ --material README.md \ --agents builder:codex,critic:claude,security-reliability:codex \ | tee /tmp/ai-meeting-create.json \ | node -e 'let s=""; process.stdin.on("data", d => s += d); process.stdin.on("end", () => console.log(JSON.parse(s).meetingDir));') node ai-meeting/scripts/ai-meeting.mjs round \ --meeting-dir "$MEETING_DIR" \ --round 1 \ --dry-run sed -n '1,220p' "$MEETING_DIR/dry-run/round-1/builder.codex.prompt.md"

--dry-runwrites preview prompts underdry-run/and does not mutate official round state.

Example 4: Continue A Meeting With A Third Round

node ai-meeting/scripts/ai-meeting.mjs round --meeting-dir "$MEETING_DIR" --round 1 node ai-meeting/scripts/ai-meeting.mjs round --meeting-dir "$MEETING_DIR" --round 2 # Add one more cross-questioning round if the first two rounds expose unresolved disagreement. node ai-meeting/scripts/ai-meeting.mjs round --meeting-dir "$MEETING_DIR" --round 3 --max-rounds 5 node ai-meeting/scripts/ai-meeting.mjs synthesize --meeting-dir "$MEETING_DIR"

Example 5: Use Claude As The Final Judge

node ai-meeting/scripts/ai-meeting.mjs synthesize \ --meeting-dir "$MEETING_DIR" \ --provider claude

meetings/<id>/ brief.md materials/ state.json rounds/ synthesis/ round-1-summary.md round-2-summary.md final.md

If synthesis returns an invalid report, the orchestrator writessynthesis/final.draft.md, records missing sections instate.json, and does not write officialfinal.md.

- Child agents cannot recursively runai-meeting.
- Codex defaults to read-only sandbox configuration.
- Claude Code defaults to--safe-modeand no tools.
- qoderclicn runs with no tools, an empty MCP config, and user settings only.
- Child agent cwd is an agent-scoped isolated external cache workspace derived from the meeting path, not the project root or meeting directory.
- Creating a meeting resets the external child workspace cache for that meeting path to avoid stale provider cwd state.
- Briefs, materials, peer outputs, and model outputs are treated as untrusted data.
- state.jsonis written atomically.
- Official artifacts are written with owner-only0600file permissions.
- Session IDs are redacted in stdout and omitted from final report provenance.
- Provider availability is decided by each adapter'sdoctorchecks. New agent tools are welcome through provider adapters, but adapters must report auth, prompt transport, session handling, and permission boundaries honestly.

node --check ai-meeting/scripts/ai-meeting.mjs node --test tests/ai-meeting.test.mjs python3 "${CODEX_HOME:-$HOME/.codex}/skills/.system/skill-creator/scripts/quick_validate.py" ai-meeting node ai-meeting/scripts/ai-meeting.mjs doctor --json

This is a web browser that enables your coding agent, such as Claude Code, to visit websites on your behalf and assist you in identifying bugs or creating UI test cases.

Product team workspace that makes your team's direction, priorities, and conviction persistent and accessible to your AI client via MCP.

Access fathom ai api endpoints via mcp server and llm.

Connect your Limitless Pendant data to Claude and other LLMs using the Limitless API.

Official MCP server for Buildable AI-powered development platform. Enables AI assistants to manage tasks, track progress, get project context, and collaborate with humans on software projects.

Hebbian memory for AI agents — learns file access patterns, builds neural pathways, predicts next tools/files, saves tokens

Context optimization for AI coding assistants — 99% token savings via 14 content-aware summarizers, 3-layer search, and progressive disclosure. No LLM dependency.

Give your agent persistent identity, real-time intelligence feeds, and the ability to publish and collaborate on shared feeds with other agents. Zero config, 16 tools.

Flatten Claude Code sessions: keep every prompt and event verbatim, resume at a lower token count.