ai-config-to-project
Skill65 starsChanged 2 months agoApache-2.0View on GitHub
Adapts a governed project's architecture rules, config, overrides, and AGENTS.md after ai-eng init. Manual only. Writes a proposal page and stops until the person says yes, then the person applies it manually outside the governed agent session. Trigger for "/ai-config-to-project", "adapt this project", "the init rules are wrong for this repo".
In this repository41 other files it gives its agents.
- AGENTS.mdAGENTS.md
- a11y-auditSkill
- ai-adversarial-loopSkill
- ai-agents-mdSkill
- ai-architectSkill
- ai-audit-codeSkill
- ai-audit-designSkill
- ai-audit-docsSkill
- ai-brainstormSkill
- ai-codegraphSkill
- ai-debugSkill
- ai-design-md-plannerSkill
- ai-exploreSkill
- ai-git-cleanupSkill
- ai-issue-reportSkill
- ai-noteSkill
- ai-orchestratorSkill
- ai-prototypeSkill
- ai-prSkill
- ai-read-docsSkill
- ai-researchSkill
- ai-review-uiSkill
- ai-securitySkill
- ai-stress-testSkill
- ai-test-plannerSkill
- ai-triageSkill
- ai-verifySkill
- ai-visual-recapSkill
- ai-writeSkill
- ai-writing-behaviorSkill
- build-checkSkill
- code-auditSkill
- design-checkSkill
- feature-verifySkill
- full-reviewSkill
- perf-auditSkill
- review-routerSkill
- second-opinionSkill
- security-auditSkill
- spec-extractSkill
- verify-featureSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 4 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 14
If the person did not invoke this skill by name, stop.
- Distinctive: line 20
Read the tree. Propose. Stop.
- Distinctive: line 22
Do not Edit or Write these files yourself.
- Very specific: line 33
Write .ai-engineering/config-proposal.html in the artifact shell (the same tokens as templates/brainstorm.html.tpl, including the …
- Very specific: line 63
Every [[guard.off]] block in the proposal has name, reason, and until (a date).
- Distinctive: line 75
Read by: the person who reviews and manually applies the proposal Dies: when the next run rewrites the page Next: none
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Adapt this project to ai-engineering
If the person did not invoke this skill by name, stop. Do not write a file. Do not run a command.
This skill is the manual door. disable-model-invocation: true is the Claude Code latch. The sentence above is the latch for every other host. Do not claim that Cursor, Pi, or Codex honor the frontmatter field.
What you may change, and how
Read the tree. Propose. Stop. The person says approve, ok, or go. The person applies the approved changes manually outside the governed agent session.
Do not Edit or Write these files yourself. The session fence denies it, and that denial stays:
.ai-engineering/arch.rules.json.ai-engineering/config.toml.ai-engineering/overrides.toml.ai-engineering/ai-eng.lock
ai-eng adapt apply is intentionally disabled: a same-user agent can create a matching hash, so the hash cannot prove human approval. The person applies the reviewed changes manually outside the governed agent session.
The page
Write .ai-engineering/config-proposal.html in the artifact shell (the same tokens as templates/brainstorm.html.tpl, including the scroll-spy script). The page shows each diff in words a person can refuse.
The machine-readable payload is one JSON object inside the page. Keys, only these, sorted:
AGENTS.md.ai-engineering/arch.rules.json.ai-engineering/config.toml.ai-engineering/overrides.toml
<meta name="ai-eng-proposal-sha256" content="<sha256 of the script body>">
<script type="application/json" id="payload">…</script>The sha256 is the hex digest of the exact script body. < inside that body is the six characters \u003c. If you change the body after stamping the hash, apply refuses.
Survivors
Survivors that a proposal must keep: Security, Lifecycle, Anti-drift, Voice.
Security is the hook, linter, and secrets floor. Lifecycle is the spoken yes and the agent running the command. Anti-drift is the line that deletes a rule the code already states. Voice is the pointer to ai-write, not a copy of the standard.
KISS, YAGNI, DRY, SOLID, and Clean Code stay only when the proposal names the check that fails when they are broken. A slogan with no check leaves.
Build and test commands come from this tree. Run them. Do not copy another repo's command line.
Layer names come from directories that exist here. A second run starts from the files on disk. It does not restore the framework's cli / guards / chain layers.
Overrides
Every [[guard.off]] block in the proposal has name, reason, and until (a date). Show that block on the page as the dangerous one. No until means you do not include the block. Apply will refuse it anyway.
After the yes
The person manually applies the approved files from outside the governed agent session. Do not run ai-eng adapt apply or pass --no-verify. If the person asks about that command, explain that it is disabled because it cannot authenticate approval.
The page uses the artifact design system in ai-brainstorm › references/artifact-design.md. Copy that document's whole CSS block and the scroll-spy script verbatim — tokens, layout rules and components, not tokens alone. Do not invent a second style.
Lifecycle
Lane: any Writes: .ai-engineering/config-proposal.html Read by: the person who reviews and manually applies the proposal Dies: when the next run rewrites the page Next: none
Source: ai-engineering (own), Apache-2.0.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

