Skip to main content
Install the skill by copying its directory into the appropriate location inside your repository or home directory. The skill directory itself is never modified by the workflow — all task artifacts are written into the target repository.

Project skill locations

Install as a project skill at:
.agents/skills/repo-task-proof-loop/
Set SKILL_PATH when running the bundled scripts:
SKILL_PATH=.agents/skills/repo-task-proof-loop

Personal skill locations

You can also install the skill once in your home directory and use it across multiple repositories.
$HOME/.agents/skills/repo-task-proof-loop/

Using both tools on the same repository

If you use Codex and Claude Code on the same repository, install the skill in both project locations or keep one canonical copy and sync it.
.agents/skills/repo-task-proof-loop/   ← Codex project skill
.claude/skills/repo-task-proof-loop/   ← Claude Code project skill
The same skill directory can be reused in either product. When you run init, it writes repo-local workflow files into the target repository — not into the skill directory — so both installations can safely share the same task artifacts under .agent/tasks/.

The SKILL_PATH variable

All bundled scripts use a path relative to the skill root. Before running any script, set SKILL_PATH to the installed skill directory:
SKILL_PATH=.agents/skills/repo-task-proof-loop
Then reference scripts through that variable:
python3 "$SKILL_PATH/scripts/task_loop.py" init --task-id my-task

Smoke test

After installing the skill, run the bundled smoke test to verify the skill structure and confirm the initialization scripts work correctly:
python3 "$SKILL_PATH/scripts/verify_package.py"
This checks the skill structure, initializes temporary repositories, installs the task artifacts and subagents, and verifies the generated task bundles and guide behavior.

After install: Claude Code session note

In Claude Code, if init creates or refreshes .claude/agents/* during a running session, start a new Claude Code session before expecting those updated agents to appear. Use /agents to inspect the available Claude agents after the new session starts.
This applies to the initial install as well as any subsequent init run that refreshes the agent files.

Build docs developers (and LLMs) love