4th-layer-agent-harness

name	4th-layer-agent-harness
description	Installs the 4th Layer Engineering harness into a repository. Claude Code and Codex hooks, run tracking, multi-agent council protocol, subagent guides, handoff templates with memory promotion, and CLAUDE.md/AGENTS.md workflow sections. Use when setting up a new repo for agent-driven development, or when adding structured run tracking and multi-agent workflow support to an existing project.

4th Layer Agent Harness — Engineering Workflow

This skill installs the 4th Layer Engineering workflow infrastructure into the current repository. It takes ~10 minutes and leaves you with a complete agent operating environment: run tracking, subagent protocol, hooks, templates, and Claude/Codex workflow sections.

---

Prerequisites

Before proceeding, confirm:

• You are in the root directory of the target repository
• Git is initialized (git init if not)
• Python 3 is available (for hooks — they use stdlib only)
• ~/bin/agent is set up if you plan to use Cursor as a subagent (see SETUP.md)

Skill package location: This SKILL.md and all accompanying files are in 4th-layer-scaffold/ at the Code Projects root. All files/ paths below are relative to that directory.

---

What Gets Installed

Artifact	Destination in Target Repo	Purpose
7 hook scripts	.claude/hooks/	Agent type enforcement, git safety, session continuity, subagent protocol, TASK_LOG reminders
Hook settings template	.claude/settings-hooks-template.json	Hooks wiring to merge into settings.local.json
3 Codex hook scripts	.codex/hooks/	Git safety, session continuity, TASK_LOG reminders
Codex hooks template	.codex/hooks.json	Codex hook wiring
Codex hooks config template	.codex/config-hooks-template.toml	Enables Codex hooks via [features] codex_hooks = true
Shared hook logic	.agent-harness/hooks/	Runtime-neutral Python logic imported by Claude and Codex hook adapters
CLAUDE.local.md template	Project root	Active Tasks table + Agent Voice (gitignored runtime state)
AGENTS.local.md template	Project root	Active Tasks table + Codex local notes read by the harness SessionStart hook (gitignored runtime state)
ARCHIVE.template.md (tracked) + ARCHIVE.md (gitignored)	runs/CLAUDE-RUNS/	Template stays tracked as the blank canonical form; ARCHIVE.md is the per-developer working log (gitignored so multi-dev archive histories don't conflict).
skills/README.md	Optional project-local skills/ or .agents/skills/	Scaffold for domain skills when the user explicitly requests repo-specific skills
skills/task-tracking/	Global by default; project-local only on request	HIGHLY RECOMMENDED. Run tracking protocol, templates (TASK_LOG, SPEC, HANDOFF, FINDINGS), init-run.sh, docstring validation
skills/context-loading/	Global by default; project-local only on request	HIGHLY RECOMMENDED. Minimal file/doc loading strategy, partial reads, cached summaries, composed shell command patterns
skills/docstring-guide/	Global by default; project-local only on request	HIGHLY RECOMMENDED. Load before any task that writes or edits actual code; keeps docstrings, navigation refs, and test traceability accurate
skills/archive-run/	Global by default; project-local only on request	HIGHLY RECOMMENDED. Archival checklist, metric collectors, ARCHIVE.md entry format
skills/subagent-management/	Global by default; project-local only on request	HIGHLY RECOMMENDED. Subagent delegation decisions, prompt structure, run-directory protocol, FINDINGS.md integration
skills/cursor-subagent/SKILL.md	Global by default; project-local only on request	Cursor agent subagent invocation (Windows/Git Bash)
skills/codex-subagent/SKILL.md	Global by default; project-local only on request	Codex CLI subagent invocation (Windows/Git Bash)
skills/gemini-subagent/SKILL.md	Global by default; project-local only on request	Gemini CLI subagent invocation
skills/adversarial-review/SKILL.md	Global by default; project-local only on request	Adversarial review protocol
skills/council-guide/SKILL.md	Global by default; project-local only on request	Quad-model council protocol
snippets/CLAUDE_MD_WORKFLOW.md	snippets/ (in target repo)	Slim workflow pointer for CLAUDE.md (full protocol lives in task-tracking skill)
snippets/AGENTS_MD_WORKFLOW.md	snippets/ (in target repo)	Slim workflow pointer for Codex AGENTS.md
Optional addons	optional/ (pick what applies)	Language/CI-specific hooks
↳ optional/python/ (2 hooks)	.claude/hooks/	Post-edit length check + ruff/black linting (Python only)
↳ optional/github-actions/ (1 hook)	.claude/hooks/	SHA-pinning enforcement for CI workflows
↳ optional/spec-quality-gate.py	.claude/hooks/	SPEC quality checklist on write (for council-style repos)

Install skills globally by default: ~/.claude/skills/<name>/ for Claude Code and ~/.agents/skills/<name>/ for Codex. Only copy skills into the target repo when the user explicitly asks for project-specific skills. The same files/skills/<name>/ payload is used for both Claude and Codex.

---

Installation

Work through these steps in order. Do not skip steps.

---

Step 1: Create directories

Create all required directories in the target repo:

mkdir -p .claude/hooks
mkdir -p .codex/hooks
mkdir -p .agent-harness/hooks
mkdir -p runs/CLAUDE-RUNS
mkdir -p snippets

Do not create skills/ or .agents/skills/ in the target repo unless the user explicitly requests project-specific skills.

---

Step 2: Copy shared hook support

The Claude and Codex hook files are thin runtime-specific adapters. Copy the shared Python hook logic first:

Source (in skill package)	Destination (in target repo)
files/.agent-harness/hooks/__init__.py	.agent-harness/hooks/__init__.py
files/.agent-harness/hooks/git_safety.py	.agent-harness/hooks/git_safety.py
files/.agent-harness/hooks/run_state.py	.agent-harness/hooks/run_state.py
files/.agent-harness/hooks/task_log.py	.agent-harness/hooks/task_log.py

Do not put shared hook logic under .claude/ or .codex/; those directories hold runtime-specific adapters only.

---

Step 2b: Copy Claude Code hook scripts

Read each file from the skill package and write it to the target repo:

Source (in skill package)	Destination (in target repo)
files/.claude/hooks/force-general-purpose-agent.py	.claude/hooks/force-general-purpose-agent.py
files/.claude/hooks/block-destructive-git.py	.claude/hooks/block-destructive-git.py
files/.claude/hooks/session-run-prompt.py	.claude/hooks/session-run-prompt.py
files/.claude/hooks/subagent-context.py	.claude/hooks/subagent-context.py
files/.claude/hooks/subagent-directory-protocol.py	.claude/hooks/subagent-directory-protocol.py
files/.claude/hooks/plan-mode-reminder.py	.claude/hooks/plan-mode-reminder.py
files/.claude/hooks/task-log-reminder.py	.claude/hooks/task-log-reminder.py

---

Step 3: Copy settings template

Read files/.claude/settings-hooks-template.json from the skill package and write it to .claude/settings-hooks-template.json in the target repo.

This contains only the "hooks" section. The user will merge it into their settings.local.json (Step 10).

---

Step 3b: Copy Codex hook scripts and templates

If the target repo will be used with Codex, copy these files:

Source (in skill package)	Destination (in target repo)
files/.codex/hooks/block-destructive-git.py	.codex/hooks/block-destructive-git.py
files/.codex/hooks/session-run-prompt.py	.codex/hooks/session-run-prompt.py
files/.codex/hooks/task-log-reminder.py	.codex/hooks/task-log-reminder.py
files/.codex/hooks.json	.codex/hooks.json
files/.codex/config-hooks-template.toml	.codex/config-hooks-template.toml

Codex hooks are behind a feature flag. In the target repo, merge the template's contents into .codex/config.toml or another active Codex config layer:

[features]
codex_hooks = true

If .codex/config.toml already exists, do not overwrite it. Merge only the [features] value.

Codex hooks are guardrails, not a security boundary. Keep Codex sandboxing and approval settings in place.

Project-local Codex hooks load only when the project .codex/ config layer is trusted by Codex. If Codex prompts about trusting project config, trust it only after reviewing the hook scripts and config.

---

Step 4: Copy run infrastructure

Source	Destination
files/runs/.gitkeep	runs/.gitkeep (empty file)
files/runs/CLAUDE-RUNS/.gitkeep	runs/CLAUDE-RUNS/.gitkeep (empty file)
files/runs/CLAUDE-RUNS/ARCHIVE.template.md	runs/CLAUDE-RUNS/ARCHIVE.template.md (tracked)
files/runs/CLAUDE-RUNS/ARCHIVE.template.md	runs/CLAUDE-RUNS/ARCHIVE.md (copy of template; becomes the dev's local working log — gitignored)

Why the split: ARCHIVE.template.md is the scaffold-canonical blank form — tracked, shared, rarely changes. ARCHIVE.md is the per-developer archive history — gitignored so each dev can archive runs on their machine without creating merge conflicts or leaking private run history into the shared repo.

Note: init-run.sh and all run templates (TASK_LOG, SPEC, HANDOFF, FINDINGS) now live inside the task-tracking skill and are installed in Step 7. The skill's scripts/init-run.sh resolves templates from ../templates/ relative to itself.

---

Step 5: Copy local runtime state template

CLAUDE.local.md is per-machine runtime state. It belongs in the target repo root, but it must stay gitignored and should not be committed.

Source	Destination	Runtime
files/CLAUDE.local.md	CLAUDE.local.md	Claude Code
files/AGENTS.local.md	AGENTS.local.md	Codex

If CLAUDE.local.md does not already exist in the target repo, copy the template from files/CLAUDE.local.md.

If CLAUDE.local.md already exists, do not overwrite it. Verify it has the Active Tasks table, and merge only the missing table/template pieces if needed while preserving any existing Agent Voice or per-machine notes.

For Codex, do the same with AGENTS.local.md: copy files/AGENTS.local.md only if the target repo does not already have one. Codex does not auto-load this file; the 4th Layer Codex SessionStart hook reads it when hooks are enabled.

---

Step 6: Documentation now lives in skills

Do not copy duplicate docs/coding_agents/ versions of agent workflow guides. The harness keeps those protocols in skills so Claude Code and Codex can load them on demand:

• context-loading replaces MEMORY_OPTIMIZATION.md
• docstring-guide replaces DOCSTRING_GUIDE.md
• subagent-management replaces SUBAGENT_GUIDE.md
• gemini-subagent replaces GEMINI_CLI_SUBAGENT.md
• adversarial-review replaces ADVERSARIAL_REVIEW.md
• council-guide replaces COUNCIL_GUIDE.md

Note: docstring_validation_template.md also lives inside the task-tracking skill at templates/docstring_validation_template.md and is installed in Step 7.

---

Step 7: Install skills

This harness ships 10 skills. Install all skills globally by default for the agent runtime(s) being configured. Project-local copies are only for users who explicitly ask for repo-specific skills.

Core skills:

• task-tracking — run tracking protocol, templates, init-run.sh, docstring validation
• context-loading — minimal file/doc loading, partial reads, cached summaries, composed shell commands
• docstring-guide — load before any task that writes or edits actual code; AI-friendly docstring navigation, file::symbol refs, stale-comment cleanup
• archive-run — archival checklist, metric collectors, ARCHIVE.md entry format
• subagent-management — delegation decisions, subagent prompts, directories, FINDINGS.md integration

Invocation and review skills:

• cursor-subagent — invocation patterns for Cursor's agent CLI
• codex-subagent — invocation patterns for Codex CLI
• gemini-subagent — invocation patterns for Gemini CLI
• adversarial-review — skeptical-but-fair multi-model review protocol
• council-guide — quad-model council deliberation protocol

Default: install skills globally for the agent runtime(s) being configured. Do not add project-local skills/ or .agents/skills/ directories unless the user explicitly asks for repo-specific skills.

Ask the user before proceeding:

"I'll install the harness skills globally by default.

A) Global - copied to the relevant global skills directory: ~/.claude/skills/<name>/ for Claude Code and/or ~/.agents/skills/<name>/ for Codex.

B) Project-local too - also copy to the active agent's project-local skills directory (skills/<name>/ for Claude Code, .agents/skills/<name>/ for Codex). Choose this only if you want repo-specific skill copies."

Then execute the chosen option:

For Claude Code, use ~/.claude/skills/ for global install. Use skills/ only if the user explicitly requests project-local Claude skills.

For Codex, use ~/.agents/skills/ for global install. Use .agents/skills/ only if the user explicitly requests project-local Codex skills. Codex does not use ~/.claude/skills/, CLAUDE.md, or CLAUDE.local.md automatically.

A — Claude Code global install:

mkdir -p ~/.claude/skills/task-tracking/{templates,scripts}
mkdir -p ~/.claude/skills/context-loading/agents
mkdir -p ~/.claude/skills/docstring-guide/{agents,references}
mkdir -p ~/.claude/skills/archive-run/scripts/metrics
mkdir -p ~/.claude/skills/subagent-management/agents
mkdir -p ~/.claude/skills/cursor-subagent
mkdir -p ~/.claude/skills/codex-subagent
mkdir -p ~/.claude/skills/gemini-subagent
mkdir -p ~/.claude/skills/adversarial-review
mkdir -p ~/.claude/skills/council-guide

For task-tracking, copy the entire skill directory:

• files/skills/task-tracking/SKILL.md → ~/.claude/skills/task-tracking/SKILL.md
• files/skills/task-tracking/templates/* → ~/.claude/skills/task-tracking/templates/
• files/skills/task-tracking/scripts/init-run.sh → ~/.claude/skills/task-tracking/scripts/init-run.sh
• chmod +x ~/.claude/skills/task-tracking/scripts/init-run.sh

For context-loading, copy the entire skill directory:

• files/skills/context-loading/SKILL.md → ~/.claude/skills/context-loading/SKILL.md
• files/skills/context-loading/agents/openai.yaml → ~/.claude/skills/context-loading/agents/openai.yaml

For docstring-guide, copy the entire skill directory:

• files/skills/docstring-guide/SKILL.md → ~/.claude/skills/docstring-guide/SKILL.md
• files/skills/docstring-guide/agents/openai.yaml → ~/.claude/skills/docstring-guide/agents/openai.yaml
• files/skills/docstring-guide/references/DOCSTRING_GUIDE.md → ~/.claude/skills/docstring-guide/references/DOCSTRING_GUIDE.md

For archive-run, copy the entire skill directory:

• files/skills/archive-run/SKILL.md → ~/.claude/skills/archive-run/SKILL.md
• files/skills/archive-run/scripts/* → ~/.claude/skills/archive-run/scripts/
• chmod +x ~/.claude/skills/archive-run/scripts/archive-run.sh
• chmod +x ~/.claude/skills/archive-run/scripts/metrics/*.sh

For subagent-management, copy the entire skill directory:

• files/skills/subagent-management/SKILL.md → ~/.claude/skills/subagent-management/SKILL.md
• files/skills/subagent-management/agents/openai.yaml → ~/.claude/skills/subagent-management/agents/openai.yaml

For lightweight invocation/review skills, copy each files/skills/<name>/SKILL.md → ~/.claude/skills/<name>/SKILL.md.

B — Claude Code project-local install, only if explicitly requested:

Source	Destination
files/skills/README.md	skills/README.md
files/skills/task-tracking/ (entire dir)	skills/task-tracking/
files/skills/context-loading/ (entire dir)	skills/context-loading/
files/skills/docstring-guide/ (entire dir)	skills/docstring-guide/
files/skills/archive-run/ (entire dir)	skills/archive-run/
files/skills/subagent-management/ (entire dir)	skills/subagent-management/
files/skills/cursor-subagent/SKILL.md	skills/cursor-subagent/SKILL.md
files/skills/codex-subagent/SKILL.md	skills/codex-subagent/SKILL.md
files/skills/gemini-subagent/SKILL.md	skills/gemini-subagent/SKILL.md
files/skills/adversarial-review/SKILL.md	skills/adversarial-review/SKILL.md
files/skills/council-guide/SKILL.md	skills/council-guide/SKILL.md

After copying, make scripts executable:

chmod +x skills/task-tracking/scripts/init-run.sh
chmod +x skills/archive-run/scripts/archive-run.sh
chmod +x skills/archive-run/scripts/metrics/*.sh

A — Codex global install:

Create the Codex global skill directories and copy the same skill payload to ~/.agents/skills/<name>/:

mkdir -p ~/.agents/skills/task-tracking/{templates,scripts}
mkdir -p ~/.agents/skills/context-loading/agents
mkdir -p ~/.agents/skills/docstring-guide/{agents,references}
mkdir -p ~/.agents/skills/archive-run/scripts/metrics
mkdir -p ~/.agents/skills/subagent-management/agents
mkdir -p ~/.agents/skills/cursor-subagent
mkdir -p ~/.agents/skills/codex-subagent
mkdir -p ~/.agents/skills/gemini-subagent
mkdir -p ~/.agents/skills/adversarial-review
mkdir -p ~/.agents/skills/council-guide

Use the same source files listed in the Claude Code global install section, but replace ~/.claude/skills/ with ~/.agents/skills/.

Restart Codex if newly installed skills do not appear.

B — Codex project-local install, only if explicitly requested:

Copy the same skill payload into .agents/skills/:

Source	Destination
files/skills/task-tracking/ (entire dir)	.agents/skills/task-tracking/
files/skills/context-loading/ (entire dir)	.agents/skills/context-loading/
files/skills/docstring-guide/ (entire dir)	.agents/skills/docstring-guide/
files/skills/archive-run/ (entire dir)	.agents/skills/archive-run/
files/skills/subagent-management/ (entire dir)	.agents/skills/subagent-management/
files/skills/cursor-subagent/SKILL.md	.agents/skills/cursor-subagent/SKILL.md
files/skills/codex-subagent/SKILL.md	.agents/skills/codex-subagent/SKILL.md
files/skills/gemini-subagent/SKILL.md	.agents/skills/gemini-subagent/SKILL.md
files/skills/adversarial-review/SKILL.md	.agents/skills/adversarial-review/SKILL.md
files/skills/council-guide/SKILL.md	.agents/skills/council-guide/SKILL.md

Make scripts executable:

chmod +x .agents/skills/task-tracking/scripts/init-run.sh
chmod +x .agents/skills/archive-run/scripts/archive-run.sh
chmod +x .agents/skills/archive-run/scripts/metrics/*.sh

Codex compatibility note: The skill format is shared. The existing files/skills/<name>/SKILL.md files are valid Codex skills because they use the standard name and description frontmatter plus Markdown instructions.

---

Step 7b: Configure archive metrics

Ask the user:

"What metrics should be collected when archiving a run?

I'll create .claude/archive-run.config with the collectors you want. The git collector (files changed, lines +/-, commits) is always included.

What applies to this project?

1. Python — pytest summary, ruff errors, mypy errors, coverage %
2. Node.js — eslint counts, jest summary
3. Go — go test, govet
4. Rust — cargo test, clippy
5. Code duplication (jscpd) — run manually with repo-specific ignore patterns
6. None of the above — git metrics only"

Based on the answer, create .claude/archive-run.config with the appropriate collectors uncommented. Use files/skills/archive-run/scripts/archive-run.config.example as the base.

If the user isn't sure yet, copy the example file as-is (git-only default) and note they can uncomment collectors later.

---

Step 8: Copy snippets

Source	Destination
files/../snippets/CLAUDE_MD_WORKFLOW.md	snippets/CLAUDE_MD_WORKFLOW.md
files/../snippets/AGENTS_MD_WORKFLOW.md	snippets/AGENTS_MD_WORKFLOW.md

(The snippets directory is at 4th-layer-scaffold/snippets/, not inside files/.)

---

Step 9: Update .gitignore

Append the following to .gitignore (create the file if it doesn't exist):

# Claude agent run directories (local only — contents never committed)
runs/
!runs/.gitkeep
!runs/CLAUDE-RUNS/.gitkeep
!runs/CLAUDE-RUNS/ARCHIVE.template.md

# Per-machine Claude runtime state
/CLAUDE.local.md

# Per-machine Codex runtime state
/AGENTS.local.md

# Log files
*.log

Do NOT replace an existing .gitignore — append only. Do NOT force-add CLAUDE.local.md. If the target repo already tracks CLAUDE.local.md, ask the user before running git rm --cached CLAUDE.local.md so the file remains locally but is removed from version control.

Do the same for AGENTS.local.md if it already exists and is tracked.

---

Step 10: Force-add tracked files

git add -f runs/.gitkeep
git add -f runs/CLAUDE-RUNS/.gitkeep
git add -f runs/CLAUDE-RUNS/ARCHIVE.template.md
# NOTE: Do NOT force-add runs/CLAUDE-RUNS/ARCHIVE.md — it is intentionally
# gitignored as a per-developer archive log.

Note: init-run.sh is no longer at runs/CLAUDE-RUNS/ — it lives inside the task-tracking skill directory (installed in Step 7). The skill's scripts/init-run.sh was already made executable in Step 7.

---

Step 11: Test run infrastructure

Run init-run.sh from the global skill install by default:

# Claude Code global default:
~/.claude/skills/task-tracking/scripts/init-run.sh test-tent
# Codex global default:
~/.agents/skills/task-tracking/scripts/init-run.sh test-tent

If the user explicitly requested project-local skills, use the project-local path for that runtime instead:

skills/task-tracking/scripts/init-run.sh test-tent
.agents/skills/task-tracking/scripts/init-run.sh test-tent

Expected output: ✅ Created: runs/CLAUDE-RUNS/RUN-YYYYMMDD-HHMM-test-tent/ with TASK_LOG.md, SPEC_v1.md, and HANDOFF.md present with timestamps substituted (not {{RUN_ID}} literals).

If template variables aren't substituted, check that the templates/ directory inside the task-tracking skill is populated correctly from Step 7.

---

Customization Pass

The infrastructure is now installed. The former docs/coding_agents/ workflow guides now live in skills, so this pass focuses on repo instruction files and any project-specific notes the user wants to add.

Customize: CLAUDE.md workflow section

The former context-loading checklist now lives in the context-loading skill, the former docstring guide now lives in the docstring-guide skill, and the former subagent guide now lives in the subagent-management skill.

The file snippets/CLAUDE_MD_WORKFLOW.md contains a slim workflow pointer section ready to paste into CLAUDE.md. The full task-tracking protocol now lives in the task-tracking skill (installed in Step 7), so CLAUDE.md only needs a pointer.

Ask the user: "Do you have an existing CLAUDE.md, or should I create a new one?"

If creating new: Copy snippets/CLAUDE_MD_WORKFLOW.md to CLAUDE.md. Then add a project-specific header section above the workflow section covering:

• Project name and purpose
• Documentation map (what docs exist and when to use them)
• Key architectural concepts
• Quick reference (ports, commands, critical file locations)
• Common pitfalls

If merging with existing: Paste the workflow sections from snippets/CLAUDE_MD_WORKFLOW.md into the appropriate place in the existing CLAUDE.md. At minimum, ensure these sections are present:

• Task Execution Protocol (pointer to task-tracking skill)
• Subagent Usage + Codex + Cursor sections
• Background Process Guidelines
• Timestamp reminder

Then ensure CLAUDE.local.md exists with the Active Tasks table (and optionally the Agent Voice section). If missing, copy files/CLAUDE.local.md to CLAUDE.local.md. This file is gitignored — it holds per-machine runtime state.

After merging, fill in all [TODO: project-specific] markers in CLAUDE.md.

Customize: AGENTS.md workflow section

If the repo will be used with Codex, use snippets/AGENTS_MD_WORKFLOW.md.

If no AGENTS.md exists: Copy snippets/AGENTS_MD_WORKFLOW.md to AGENTS.md. Then add project-specific repository guidance above the workflow section.

If AGENTS.md already exists: Merge the workflow sections from snippets/AGENTS_MD_WORKFLOW.md into the existing file.

Then ensure AGENTS.local.md exists with the Active Tasks table. If missing, copy files/AGENTS.local.md to AGENTS.local.md. This file is gitignored.

---

Finalize: Configure settings.local.json

Merge .claude/settings-hooks-template.json into .claude/settings.local.json:

• If no settings.local.json exists: rename the template to settings.local.json
• If it exists: merge the "hooks" key into the existing file without overwriting other keys (permissions, enabledMcpjsonServers, additionalDirectories, etc.)

After merging, delete .claude/settings-hooks-template.json.

Finalize: Configure Codex hooks

If using Codex, ensure .codex/hooks.json exists and .codex/config.toml enables hooks:

[features]
codex_hooks = true

If .codex/config.toml already exists, merge the feature flag without overwriting other Codex settings.

---

Completion Checklist

Verify everything before starting work:

• task-tracking, context-loading, docstring-guide, archive-run, and subagent-management skills installed globally, unless the user explicitly requested project-local skill copies
• init-run.sh test-tent generates TASK_LOG.md, SPEC_v1.md, and HANDOFF.md with substituted timestamps
• .agent-harness/hooks/ exists with shared hook logic
• .claude/settings.local.json has "hooks" key with 7 core hooks + SessionStart
• .codex/hooks.json has SessionStart, UserPromptSubmit, and PreToolUse hook wiring (if using Codex)
• .codex/config.toml or another active Codex config layer has [features] codex_hooks = true (if using Codex)
• .claude/archive-run.config created with appropriate metric collectors
• context-loading skill installed and referenced by CLAUDE.md/AGENTS.md workflow section
• docstring-guide skill installed and referenced by CLAUDE.md/AGENTS.md workflow section
• subagent-management skill installed and referenced by CLAUDE.md/AGENTS.md workflow section
• CLAUDE.md has the workflow pointer section (skill references, subagent section)
• AGENTS.md has the workflow pointer section (if using Codex)
• CLAUDE.local.md has the Active Tasks table (and Agent Voice if used)
• AGENTS.local.md has the Active Tasks table (if using Codex)
• .gitignore has runs/, !runs/.gitkeep, /CLAUDE.local.md, and /AGENTS.local.md appended
• CLAUDE.local.md exists locally but is not tracked by git
• AGENTS.local.md exists locally but is not tracked by git (if using Codex)
• runs/.gitkeep, runs/CLAUDE-RUNS/.gitkeep, ARCHIVE.template.md force-added to git
• runs/CLAUDE-RUNS/ARCHIVE.md exists locally and is not tracked

---

What's NOT installed (by design)

The following were intentionally excluded. Add them yourself if needed:

What	Why excluded	Available as optional addon?
post-edit-lint.py hook	Python/ruff/black specific	Yes — optional/python/post-edit-lint.py
post-edit-length-check.py hook	Python AST specific	Yes — optional/python/post-edit-length-check.py
github-actions-security.py hook	GitHub Actions specific	Yes — optional/github-actions/github-actions-security.py
spec-quality-gate.py hook	Opinionated for council-style repos	Yes — optional/spec-quality-gate.py
Project-specific permissions in settings.local.json	You know your project's allowed bash commands.	No
Full AGENTS.md	Project-specific repository guidelines. Scaffold only provides a workflow snippet.	Snippet only — snippets/AGENTS_MD_WORKFLOW.md
EXAMPLES.md, VISUAL_DIAGRAMS.md	Project-specific code examples. Write your own.	No
GEMINI-RUNS/	Add runs/GEMINI-RUNS/ yourself if you use Gemini agents.	No
MCP query templates	Specific to your database tooling.	No
Project-tailored DOCSTRING_GUIDE.md	The reusable guide lives in the docstring-guide skill; copy/tailor a repo doc only on explicit request.	Skill: docstring-guide
ARCHITECTURE_CONSTRAINTS.md	Project-specific hard rules. Write your own.	No (TODO template planned)
health-pulse.sh	Language-specific metrics gathering. Now partially covered by archive-run skill's metric collectors.	Partially — see archive-run skill