Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →ai-builder-club/setup-codebase-harness
Master skill — set up the full agent harness for any repo so an agent can work it reliably: legible (map-not-manual docs + custom lints), executable (one-command dev stack), verifiable (e2e gate + a verify-before-ship loop), plus commit hygiene and entropy control.
Passed 15 of 18 test cases with the skill, and 11 of 18 without it.
and 4 more in the full report
6 fixed12 passed either way3 still failing1 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 6 test cases the skill fixed. Same model, same prompt, with and without the skill.
To enable AI agents to run our full application stack locally, we need to standardize the local environment startup command script. What standard script path and subcommand format should we standardize on for local dev stack management?
| Checked | Without skill | With skill |
|---|---|---|
| Recommends standardizing the local stack launcher at scripts/dev-local.sh. | Missed | Passed |
| Specifies using the up subcommand to launch local services. | Passed | Passed |
The instructions your model reads when the skill loads.
Harness engineering: the model is fixed — what you engineer is the scaffolding around it (the environment, the docs, the feedback loops) so an agent can build and verify software with minimal human attention. Humans steer; agents execute. Your job is to make the repo legible, executable, and verifiable.
Work incrementally and depth-first: assess what exists, build the one missing capability, use it to unlock the next. Don't boil the ocean — set up what the repo actually needs. When the agent struggles, the fix is almost never "try harder" — ask "what capability is missing, and how do I make it legible and enforceable?" and add it.
This skill orchestrates the focused sub-skills: dev-local-setup,
e2e-setup, crabbox-setup (cloud/parallel), and verifier-setup
(scaffolds a repo-specific /verify loop; supersedes the older pr skill).
Survey the repo: stack, package manager, services/ports, infra deps, existing docs/tests/CI, and the implicit rules (buried in READMEs, PR comments, people's heads). Note what's missing per pillar below.
What the agent can't see doesn't exist. Knowledge in chat threads / heads is invisible — push it into versioned, repo-local artifacts.
AGENTS.md / CLAUDE.md) to a
~100-line table of contents: one-line overview, project tree, golden rules
(the hard invariants), and a "where to look" table. Move the depth into a
structured docs/ system-of-record (architecture, frontend, testing, domain
topics) with a docs/index.md. A monolithic instruction file rots and crowds out
the task — keep the map small and stable, disclose detail progressively.any,
forbidden imports, file-size, structured logging). Write the error message to
inject the fix ("X isn't allowed here — do Y") so the remediation lands in agent
context. Wire them into the repo's linter + CI.test cases passed, vs 11 / 18 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.