meta-improve — Claude Code skill: what it does, how to install
A Claude Code skill applies operator-approved Dream candidates through a gated change process. Run it with /meta-improve-vault before building the change.
MIT ★ 26 updated 2026-11-16
npx skills add wrg32786/aigent-os --skill meta-improve -a claude-codeWhat meta-improve does
meta-improve applies the smallest change that meets a candidate marked approved in vault/memory/DREAM_LOG.md. Before the build, run node daemons/nightly-contracts.mjs meta --root <aigent-root> --candidate <candidate-id> and require META_GATE PASS. The gate checks the exact status and non-placeholder evidence of operator approval.
The skill reads only the evidence references named by the candidate in LESSONS.jsonl and NIGHTLY_CAPTURE_CANDIDATES.jsonl. It runs proof before and after the change, then prepares a review artifact with the diff, commands, exit codes, residual risks, and rollback. The operator receives the accept, reject, or revise choices.
It does not approve candidates, edit Dream, push, merge, or deploy. Node.js and access to the repository root and referenced files are required for the gate.
What it helps with
- apply an operator-approved Dream candidate
- gate a build on candidate status and approval evidence
- run checks before and after a proposed change
- prepare a diff and rollback details for review
Example prompts for the agent
- Run
/meta-improve-vaultfor the approved Dream candidate and prepare its change for review. - Apply the operator-approved proposal from
DREAM_LOG.mdand show the diff, checks, and rollback.
How to install meta-improve
1Skills CLI (Vercel)
npx skills add wrg32786/aigent-os --skill meta-improve -a claude-code2Manually: put the files in place
~/.claude/skills/meta-improve/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/meta-improve/
1Skills CLI (Vercel)
npx skills add wrg32786/aigent-os --skill meta-improve -a codex2Manually: put the files in place
.agents/skills/meta-improve/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add wrg32786/aigent-os --skill meta-improve -a cursor2Manually: put the files in place
.cursor/skills/meta-improve/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install wrg32786/aigent-os skills/meta-improve2Manually: put the files in place
.github/skills/meta-improve/SKILL.md1Skills CLI (Vercel)
npx skills add wrg32786/aigent-os --skill meta-improve -a gemini-cli2Platform CLI
gemini skills install https://github.com/wrg32786/aigent-os --path skills/meta-improveManually: put the files in place
.devin/skills/meta-improve/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add wrg32786/aigent-os --skill meta-improve -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about meta-improve
How do I install meta-improve in Claude Code?
Run npx skills add wrg32786/aigent-os --skill meta-improve -a claude-code. The skill file is at ~/.claude/skills/meta-improve/SKILL.md.
What does meta-improve do?
It makes a constrained change for an operator-approved Dream candidate, runs checks, and prepares review evidence. It cannot approve candidates or merge changes.
What does meta-improve need to run?
It needs Node.js, access to the repository root, and the Dream and runtime files referenced by the candidate. The gate must return META_GATE PASS before the build.
Source text
SKILL.md · MIT
--- name: meta-improve agent: none description: "Run constrained self-improvement without self-approval or self-merge. Use when implementing operator-approved Dream candidates via the canonical protocol." allowed-tools: Read, Write, Bash user-invocable: false status: PRODUCTION - invoke through /meta-improve-vault --- # Constrained self-improvement The public verb is `/meta-improve-vault`. ## Input and gate Select only from canonical `vault/memory/DREAM_LOG.md`: an exact candidate whose `status` is `approved` and whose approval was written by `/dream --set-status` from a direct operator decision. `proposed` is not approval. Read only the exact evidence references named by that approved candidate from `vault/memory/runtime/LESSONS.jsonl` and `vault/memory/runtime/NIGHTLY_CAPTURE_CANDIDATES.jsonl`. Those products provide traceability; they never provide approval by themselves. Before Build: ```text node daemons/nightly-contracts.mjs meta --root <aigent-root> \ --candidate <candidate-id> ``` Require `META_GATE PASS`. It checks the canonical block for exact approved status and non-placeholder operator evidence. This is the executable `nightly-contracts.mjs meta` gate. ## Build For one approved candidate: 1. Re-read its evidence, proposed change, mutation proof, and scope. 2. Use repository isolation where normal policy requires it. 3. Apply the smallest change satisfying the proposal. 4. Run the real red-before and green-after proof plus relevant system checks. 5. Produce a review artifact with exact diff, commands, exits, residual risks, and rollback. 6. Surface `accept / reject / revise` to the operator. Never merge, push, deploy, broaden scope, modify an operator gate, approve another candidate, or infer approval. A failed proof asks the Dream writer to set the candidate `blocked`. ## Status ownership This skill never edits Dream directly. The Dream writer owns: - direct operator approval: `proposed -> approved`; - verified build failure: `approved -> blocked`; - direct rejection: `approved -> rejected`; - independently verified merge: `approved -> merged`. The complete flow is: `propose -> operator gate -> build -> independent review -> operator outcome`
© wrg32786 · MIT · wrg32786/aigent-os