agentleFS
quaatquaat/autonomous-development/skills/autonomous-feature/SKILL.md

autonomous-feature

Skill3 starsChanged 49 days agoMITView on GitHub

Autonomously develop a repository feature from a high-level idea. Codex independently enhances the idea, proposes a detailed plan, and reviews the implementation while Claude reconciles requirements, implements, verifies, triages findings, and fixes valid issues. Use when the user delegates an end-to-end feature change.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 6 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  • could be automatic: a setting could enforce it
  1. Very specific:

    Use ultrathink for architecture, compatibility, and review triage.

    line 36
  2. Distinctive:

    Never push, merge, publish, deploy, rotate credentials, or modify remote infrastructure.

    line 43
  3. Very specific:

    Never use danger-full-access, --yolo, bypassPermissions, or equivalent unrestricted modes.

    line 44
  4. Distinctive:

    Never apply an irreversible database migration or delete user data.

    line 45
  5. Distinctive:

    Do not weaken authorization, validation, tests, or static checks to make the workflow pass.

    line 46
  6. Distinctive:

    Use no more than the configured review-round budget.

    line 48
  7. Distinctive:

    Treat every Codex finding as a proposal requiring evidence-based triage.

    line 49
  8. Distinctive:

    Do not create commits unless the user explicitly requested them.

    line 50
  9. Very specific:

    Do not declare success until evaluate succeeds:

    line 96

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Autonomous feature development

Implement this feature idea:

$ARGUMENTS

Use ultrathink for architecture, compatibility, and review triage. Operate as a state-machine driver: ask the controller what to do next, execute that phase, repeat. Detailed per-phase guidance lives in references/ and is loaded only when a phase needs it.

Non-negotiable boundaries

  • Preserve unrelated user changes.
  • Never push, merge, publish, deploy, rotate credentials, or modify remote infrastructure.
  • Never use danger-full-access, --yolo, bypassPermissions, or equivalent unrestricted modes.
  • Never apply an irreversible database migration or delete user data.
  • Do not weaken authorization, validation, tests, or static checks to make the workflow pass.
  • Codex planning and review executions must remain read-only.
  • Use no more than the configured review-round budget.
  • Treat every Codex finding as a proposal requiring evidence-based triage.
  • Do not create commits unless the user explicitly requested them.

Driver loop

  1. Confirm this is a Git repository. Inspect CLAUDE.md, repository instructions, architecture, status, and tests. Use EnterWorktree for an isolated worktree whenever available — mandatory when the starting worktree has uncommitted changes. If the user explicitly asks for current-checkout mode, stay in the current branch instead of entering a worktree, and require a clean feature branch before proceeding.
  2. Initialize:

    python3 "${CLAUDE_PLUGIN_ROOT}/scripts/controller.py" doctor
    python3 "${CLAUDE_PLUGIN_ROOT}/scripts/controller.py" init --feature "$ARGUMENTS" --mode auto --worktree-mode isolated

    init prints the run-state.json path and run ID. With multiple concurrent runs, pass --run-id <run-id> to all subsequent commands. If doctor reports a missing prerequisite, mark the run blocked and report it rather than bypassing it.

    For current-checkout mode, do not call EnterWorktree. Instead, keep the current branch checked out and run:

    python3 "${CLAUDE_PLUGIN_ROOT}/scripts/controller.py" init --feature "$ARGUMENTS" --mode auto --worktree-mode current

    The controller refuses main/master unless the user also passes --allow-main, and it refuses a dirty tree in current-checkout mode.

  3. Repeatedly ask the controller for the next phase, then execute it:

    python3 "${CLAUDE_PLUGIN_ROOT}/scripts/controller.py" next-action --json

    The response gives phase, required_action, completion_condition, and references. Read the referenced file under references/ for that phase and follow it until the completion condition holds. Phase references:

    • references/specification.md — produce the accepted spec.
    • references/planning.md — produce the accepted plan and set the risk gate.
    • references/implementation.md — implement the plan.
    • references/verification.md — run and record checks.
    • references/review.md — Codex review, triage, and adversarial review.
  4. Do not declare success until evaluate succeeds:

    python3 "${CLAUDE_PLUGIN_ROOT}/scripts/controller.py" evaluate
    python3 "${CLAUDE_PLUGIN_ROOT}/scripts/controller.py" status
    python3 "${CLAUDE_PLUGIN_ROOT}/scripts/controller.py" usage-report

Final report

  • the implemented behavior;
  • principal files changed;
  • verification commands and results;
  • Codex review rounds and disposition of findings;
  • adversarial review result when required;
  • per-phase usage table from usage-report;
  • remaining risks or explicit blocked reason;
  • a suggested conventional commit message.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.