A reusable Agent Skill for initializing, auditing, updating, maintaining, and continuously checking repositories against a lightweight agent-first harness model. Inspired by OpenAI's "Harness engineering: leveraging Codex in an agent-first world".
Compatible with Claude, Codex, OpenCode, and other agents that support the Agent Skills spec.
npx skills add github.com/kjendrzyca/repo-harness --skill repo-harnessTarget a specific agent:
npx skills add github.com/kjendrzyca/repo-harness --skill repo-harness --agent opencodeThe skill is not bootstrap-only. It supports the full lifecycle:
init- create the smallest useful local harness in a target repo.audit- assess maturity and identify gaps.update- apply one justified next-layer improvement.maintain- clean up drift and restore alignment.decide- answer whether the repo should evolve further or stay put.exec-plans- install or refresh repo-local execution-plan support.check- run a continuous conformance pass and suggest or apply the smallest useful cleanup.
Use it by asking the agent to invoke the skill:
Use the `repo-harness` skill to audit this repository.
Use the `repo-harness` skill to apply one justified next step.
Use the `repo-harness` skill to install execution plans.
The ExecPlan template in skills/repo-harness/assets/exec-plans/create-plan-file.md is based on OpenAI's execution plans guidance:
local-code-review- local Codex review skill. It runs the repo-harness review prompt from a checked-out repository, stores a Git-private continuity ledger, marks fixed prior findings as resolved, and prints the Markdown verdict. Install:npx skills add github.com/kjendrzyca/repo-harness --skill local-code-review.install-code-review-bot- private-repository-only skill that installs, authenticates, and rotates the Codex CLI PR review bot on a repo. Do not install it in public repositories. Install:npx skills add github.com/kjendrzyca/repo-harness --skill install-code-review-bot.
See AGENTS.md.