primedirective

Manual `$deliver` workflow for readable execution plans, optional deep research, optional Pro analysis, user review, and approved implementation. `--fast` skips only the initial plan-review pause. Supports `--deep-research`, `--pro-analysis`, and `--fast`.

Use when the user invokes $clarifier or asks for help clarifying a rough draft, stuck thought, post, essay, message, or explanation through a revision-coaching loop. Helps the user improve their own writing with questions, short teaching examples, and user-authored rewrites rather than ghostwriting.

Finish the current feature branch by handling dirty work safely, pushing, creating or using a PR, merging it, deleting the remote branch, switching back to the base branch, pulling the merged base, and deleting the local branch. Use when invoked with `$ship-branch`, or when the user asks to ship, finish, merge, clean up, and return to main from the current branch.

Optimize test-suite runtime and CI feedback speed through report-first audits or bounded `/goal` fix loops. Use when the user asks to speed up tests, reduce slow or flaky test feedback, profile unit/integration/e2e test runtime, improve setup or teardown cost, optimize runner or CI test execution, or reduce test harness noise while preserving validation confidence. Normal use has no mode flags.

Optimize backend and database performance through report-first audits or bounded `/goal` fix loops. Use when the user asks to speed up backend routes, perform database query sweeps, inspect schema/index health, reduce backend runtime work, or improve connection/lock/migration/observability hygiene. Normal use has no mode flags.

Optimize frontend journey speed and visual continuity through report-first audits or bounded `/goal` fix loops. Use when the user asks to improve page load speed, Core Web Vitals, interaction responsiveness, frontend bundle or asset cost, route-to-route transitions, flashes, blank gaps, layout shifts, loading-state mismatch, bfcache/back-forward behavior, or other browser-perceived performance. Normal use has no mode flags.

Convert messy source material, rough goal prompts, readable execution plans, or PRD/TDD/tasks-plan artifacts into a reviewable `tasks/goal-plan-<plan-key>.md` document that starts with the goal, done-when criteria, not-done-yet criteria, and a copy-pasteable `/goal` start prompt, then fills in the primary verifier, supporting checks, and explicit stop conditions. Use when work is an adaptive evidence loop where validation results decide the next implementation step.

Create a paste-ready prompt for ChatGPT.com Deep Research from the current thread's plan or discussion. Use when the user wants to manually run ChatGPT Deep Research with a connected GitHub repo or files, especially to get first-principles adversarial critique, current best practices, alternatives, and external research before local planning or execution.

Run a deep, adaptive, read-only analysis pass that widens the search space, uses a default adversarial council with rebuttal rounds to test competing explanations, and synthesizes the best mechanism-level answer before implementation judgment. Supports `--deep-research` for web-backed operator/current-practice research and `--pro-analysis` for ChatGPT Pro browser escalation through the repo wrapper.

Review Prime Directive skill changes before merge by running the baseline and candidate skill contracts on the same realistic scenario, preserving artifacts, and using a fresh judge to decide whether the candidate is materially better, neutral, or regressive.

Run a goal-backed current-branch merge-readiness loop. Use when invoked with `$merge-review`, especially inside `/goal $merge-review`, before a branch should be merged. Reviews `origin/main...HEAD`, records state in `tasks/merge-review-<branch-slug>.md`, fixes verified local findings, validates, and rereviews until no `Disposition: fix` findings remain or a real blocker is proven.

Run explicit branch reviews with optional task scoping. Supports ad-hoc full-branch review, task-scoped review, and `--preserve-review-artifacts`.

Review an active `$deliver` execution plan with a mandatory first-principles adversarial council loop before implementation. Defaults to applying safe plan-only fixes, supports `--approval-gate` for read-only proposed fixes, and writes `tasks/tmp/review-plan-<plan-key>.md`.

Run a full-repository sweep that always starts with a first-principles no-edit audit, then the full review chain, and pauses for approval before fixes. Invoke inside `/goal $repo-sweep` for a repair/resweep goal. Supports `--pro-analysis`, `--swarm`, `--dep-scan`, and `--preserve-review-artifacts`; `--swarm` includes nitpick-depth maintainability review by default.

Run a bounded iterative refinement loop over existing Prime Directive planning artifacts (`prd`, `tdd`, and `tasks-plan`) using one fresh reviewer subagent per critique round. Use when the user asks to improve, pressure-test, critique, iterate, loop on, or first-principles-review a plan before execution; supports `plan-key=KEY`, `--max-rounds=N`, and `--preserve-refine-artifacts`.

Bootstrap a repo's first meaningful validation surface, including lint, format, typecheck, test, build, CI, and optional local hooks. Supports `--with-hooks`.

Create or update a repo-specific `docs/ARCHITECTURE.md` that records module boundaries, dependency direction, composition roots, shared-code rules, testing boundaries, lightweight architecture checks, and accepted deviations. Use when invoked with `$create-architecture`, when bootstrapping a non-trivial repo, or before boundary-affecting work when no architecture doc exists.

Use when the user asks for a plainer, simpler, shorter, or more direct explanation, rewrite, summary, restatement, or answer. Produces concrete full-sentence explanations that lead with the main point and avoid needless jargon.

Safely clean up merged feature branches locally and on `origin` without deleting active, unmerged, or still-open work. Optionally target a specific branch.