Execute Plan Hardcore

Execute Plan (Hardcore)

Purpose

Same end-to-end plan execution as execute-plan, plus a mandatory post-execution loop: run the audit skill on the work, require A grades on all three rubric dimensions, fix everything that stands in the way, and repeat until those grades are A across the board.

Use When

• A concrete checked-in plan path exists in README/Plans.
• The user explicitly invokes $execute-plan-hardcore, use execute-plan-hardcore, or otherwise clearly asks for this workflow (not implicit routing).
• The user wants execute-plan behavior and a full audit gate with A / A / A before the work is considered done.

Do Not Use When

• No checked-in plan exists, or the request is planning-only / read-only.
• The user wants standard execute-plan without the audit-to-A loop—use execute-plan.
• The user forbids additional fix iterations after the plan is “done” by execute-plan alone.

Authority Model

• Explicit invocation of execute-plan-hardcore authorizes:

  • everything execute-plan authorizes: phase commits without pushing until all phases are done, then one sync/push/CI tail per execute-plan step 5, plus follow-up fixes;
  • after the plan is fully executed and the execute-plan tail (git-push loop) has completed as that skill describes, additional commits and changes needed to satisfy the post-audit gate below;
  • running the audit workflow as part of this skill even though the user message may not contain the substring audit—this skill is the explicit request for audit and for implementation fixes until grades reach A (overrides audit’s default “no fixes unless asked” rule for this phase only).

• If execute-plan-hardcore was only inferred or routed implicitly, stop before the first commit or push and explain why—same boundary as execute-plan.

• At workflow start, state that execute-plan-hardcore is running: phase commits through Part A, then one push/CI tail, then audit → fix → repeat until A/A/A (additional commits/pushes only as needed for that loop—not per-phase push during Part A).

Procedure

Part A — Execute the plan (same as execute-plan)

1. Read the same primers as execute-plan (plan workflow, phase-audit rubric); see also execute-plan/references/index.md.
2. Follow the execute-plan skill end-to-end: read the plan and linked guides/notes, one phase at a time, phase-audit per phase, plan checkboxes, narrow phase commits, then after all phases git-push (merge, push, CI green) per that skill.

Use the execute-plan SKILL.md as the canonical procedure for Part A.

Part B — Hardcore audit gate (after Part A is complete)

Run only when all phases are implemented, the plan reflects reality, and the execute-plan push/CI expectations are satisfied (no known failing CI from this work unless the user explicitly overrides).

1. Scope the audit to the plan execution: prefer a git-defined slice (e.g. merge base with origin/main or the branch point where this plan work started → HEAD), or the directories/files the plan touched. State the scope in the audit report header (same expectation as audit).

2. Run audit in full:

   • Mechanical signals from repo root, aligned with audit for this repo: pytest (or python -m pytest); lint only when the repo defines a standard lint command in pyproject.toml or docs—otherwise skip and note “no configured lint.” When the audited surface includes kokoro.js/, add targeted checks there (e.g. npm test) per audit.
   • Depth: solo vs parallel charters per audit (when in doubt, parallelize).
   • Produce the audit output template including Architecture, Correctness risk, and Complexity debt grades and findings.

3. Pass condition: Architecture, Correctness risk, and Complexity debt are all A. (If all three are A, overall is A per the worst-of-three rule in audit.)

4. If any of the three grades is below A:

   • Fix all issues that block raising grades to A (treat P0/P1 first; do not ignore mechanical check failures).
   • Re-run relevant checks; git-commit (and push if needed) per repo conventions so fixes are recorded.
   • Repeat from step 2 (re-scope if the diff grew) until A / A / A.

5. Stuck loop: If multiple complete audit cycles still cannot reach A / A / A due to a tradeoff that needs product or architecture buy-in, stop, report grades and the blocker, and ask the user—do not spin forever.

Worktree Rules

Same as execute-plan: ignore unrelated dirty files unless they conflict; do not revert unrelated user work.

Boundaries

• Do not skip Part B after Part A.
• Do not grade inflate; audit rubric applies.
• Do not lower audit thresholds—raise code/test quality until grades merit A.

Relation to Other Skills

• execute-plan: Part A is identical in intent; execute-plan-hardcore adds Part B only.
• audit: Part B is a full audit pass with fix iterations authorized by this skill.
• phase-audit: Still used per phase inside Part A; Part B is broader audit after the full execution lands.

Canonical Docs

Start with README/Skills/plan-workflow-skills-guide.md and README/Skills/phase-audit-rubric.md, then the active plan and linked README/Guides and README/Notes.