Adonis0123/adonis-skills / skills/workflow-gate
workflow-gate
VerifiedClassify which workflow or skill should run when route choice matters: unresolved creative work, named-option decisions, explicit Goal management, destructive actions, architecture work, planning, debugging or ship checks, review handoff, and full-completion pipelines. Skip trivial one-line work and clearly matching safe named-skill requests. Treat first-principles analysis as a method, not a Challenge trigger unless the user is widening the option space.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 13.2 KB |
| agents/openai.yaml | config | 410 B |
| evals/evals.json | config | 41.7 KB |
| evals/trigger-eval.json | config | 3.1 KB |
| references/examples-core.md | markdown | 5.2 KB |
| references/examples-edge.md | markdown | 14.9 KB |
| references/route-adjustments.md | markdown | 3.7 KB |
| references/workflow-systems.md | markdown | 9.5 KB |
references/workflow-systems.md
View on GitHub# Workflow systems — integration boundaries
> Internal classifier reference, not part of the output contract. The output remains the same 9 fields (including `Thesis`). The phase mapping and ecosystem notes below help you decide the Route — they don't surface in the emitted block except via Route / Runtime skill / Thesis.
Three workflow ecosystems show up in the runtime: `mattpocock` (grilling / architecture scan skills), `obra/superpowers` (debug / TDD / writing-plans), and `adonis-skills` (this repo: `agentic-review-handoff`, `architecture-hardening-loop`, `task-completion-loop`, `goal-gate`). `addyosmani/agent-skills` remains a phase reference only.
## Ecosystem boundaries
### Challenge + grilling — the creative-work HARD-GATE
Creative work with unresolved product/design choices and no paid design/spec routes to **Challenge**, not to a deleted `brainstorming` skill. Runtime is `grilling` (or `grill-with-docs` when the same request explicitly requires durable ADR/glossary writes).
Challenge requires a thesis:
- `Thesis: user-provided` — user already stated the preferred approach; load grilling next.
- `Thesis: agent-strawman` — draft a short strawman. Confirm when product/taste direction remains user-owned; when the user explicitly delegates the reversible direction, adopt it, record the assumption, and load grilling without another confirmation.
A request to **replicate an existing UI**, **add a new screen**, **compose a page from a design system**, or **change product/UI behavior** counts as creative work when visible behavior or product direction still has unresolved what/why choices. A fully specified component state, copy change, interaction, function, or API contract is spec-in-prompt and can use Rule #2's Light exception.
Exception: when the user references an existing design doc or spec by path, the design gate has already been paid. Route by the immediate request: **Plan** only for task breakdown, or **Light** for direct implementation regardless of scope. Add TDD when behavior has regression risk, and record the spec path in `Assumptions`.
### Named-option convergence — Challenge in convergence mode
When **named options exist and the bottleneck is picking one** (Stripe vs Lemon Squeezy, monolith vs microservices, sync vs async), stay on Challenge with `user-intent=decide`. Recommend one option as the thesis (`agent-strawman`, or `user-provided` when the user already prefers one) and pressure-test it against the named alternatives only; never widen. If the user delegates the bounded choice, decide in the same turn with `Runtime skill: none`: decisive reason plus rejected alternative, delegation recorded in `Assumptions`. A chat Decision Summary locks the choice; write a file only when the user or repository rules ask.
The retired `discuss-before-plan` skill used to own this job. Its safety semantics now live in Rule #1: a turn that can execute an irreversible action becomes an authorization hold (`Runtime skill: none`, one blocking authorization question, no work started), and a design-only comparison of future destructive options is `destructive=no; risk=high` with a mandatory Rule #1 re-gate before implementation.
### Architecture — diagnose vs harden
- `improve-codebase-architecture` (mattpocock): explore + HTML candidate report; stop after report when the user only wants diagnosis.
- `architecture-hardening-loop` (Adonis0123): bounded scan → evidence triage → fix → review by the saved Reviewer preference (default claude, codex, grok; timeout falls back to the next) → rescan. Requires an explicit user path/module **and** implement/harden intent.
Do not vendor mattpocock logic into this gate; only route to installed bare slugs.
### task-completion-loop — whole-pipeline downstream owner
`task-completion-loop` is not a Route enum. After destructive and creative safety overrides clear, hand off directly only when the user requests the whole named-plan/spec completion pipeline: work ledger, human-decision convergence, Goal ownership, implementation, agentic review, architecture hardening, and a fresh final Claude audit. Ordinary finish/test, Goal-only, review-only, and architecture-only requests stay in the normal routes.
The downstream skill owns recursive capability checks. Installed files do not prove callability: invocation-disabled dependencies, missing `codebase-design`, unavailable read-only delegation, or an unavailable real Grok/Claude session must fail closed before work starts.
### obra debug / TDD / plans — keep for Light and Plan
`systematic-debugging`, `test-driven-development`, and `writing-plans` remain the Light/Plan runtimes. Do not replace them with mattpocock skills.
### addyosmani/agent-skills — phase reference only
Partitions the lifecycle into phases (`idea-refine` / `spec-driven-development` / `planning-and-task-breakdown` / `incremental-implementation` / `code-review-and-quality` / `shipping-and-launch`) and exposes `/ship` as a parallel persona orchestrator. Useful as a mental map for where a request sits in its lifecycle — not as a `Runtime skill` source (the `agent-skills:` namespace is not resolved in the gate's single-bare-token contract today; emitting it would silently break the downstream load).
## Phase as an internal classifier
Tag the prompt's phase before committing to a Route. The Phase column is never part of the output block.
| Phase | Internal label | Best-fit Route | Runtime skill (bare slug) |
| ------------------------------------------------- | --------------- | ------------------------------------------------- | ----------------------------------------- |
| Read-only lookup | `lookup` | Direct | `none` |
| Ideation / thesis stress / creative without spec | `define-design` | Challenge | `grilling` or `grill-with-docs` |
| Named options, pick one | `decide` | Challenge (convergence) | `grilling`, or `none` when delegated |
| Spec or RFC exists, break into tasks | `plan` | Plan | `writing-plans` |
| Broad or multi-context task breakdown | `build-plan` | Plan | `writing-plans` |
| Existing-code structure diagnose | `arch-diagnose` | Architecture | `improve-codebase-architecture` |
| Scoped architecture harden loop | `arch-harden` | Architecture | `architecture-hardening-loop` |
| Sequential or context-heavy direct implementation | `build` | Light | direct local work or TDD |
| Symptom / bug / failing test | `verify-bug` | Light | `systematic-debugging` |
| Claim of done / ready-to-ship | `verify-ship` | Light | direct verification with fresh evidence |
| Cross-agent review / fix-then-re-review | `review` | Review-Handoff | `agentic-review-handoff` |
| Explicit whole named-plan completion pipeline | `full-complete` | direct downstream handoff (no Route enum) | `task-completion-loop` |
| Pre-launch checklist with persona fan-out | `ship` | Light (flag unsupported fan-out in `Assumptions`) | direct verification with available checks |
If the phase maps to a Route the gate supports, emit it. If it maps to a phase the gate does not yet model cleanly (today: persona-fan-out `ship`), route to the nearest existing Route and surface the gap in `Assumptions`.
## Invariants
- 9-field output, same order, same enums (`Thesis` included; `n/a` when not Challenge).
- `Runtime skill` is single bare token; `Fallback alias` is the only place plugin-namespaced strings appear.
- Never emit or load `brainstorming`.
- Rule #1 (destructive) overrides everything else.
- Widening / thesis stress and named-option convergence are both Challenge; `user-intent` (ideate vs decide) and `Thesis` record the mode.
- Architecture harden requires explicit path **and** implement intent.
## Native Goal ownership and runtime policy
A primary native Goal request goes directly to `goal-gate` after Rules #1/#2 clear. Creating, managing, continuing, completing, or blocking a Goal is not ordinary Plan generation, and does not alone request `task-completion-loop`. Preserve exclusions on reviewers, scanners, and execution systems.
The full-completion handoff needs the whole pipeline: one work ledger, real host + Grok + Claude convergence, Goal-managed implementation, agentic review, scoped architecture hardening, and fresh final Claude audit for a named plan/spec or bounded unfinished task. Recursive dependency checks belong to that downstream owner. Invocation policy and actual host capabilities matter as much as the presence of skill files: `disable-model-invocation: true` and `allow_implicit_invocation: false` block implicit handoffs unless the user names the dependency and the host permits it. This keeps a routing decision from silently becoming permission to simulate unavailable workflows.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deployment | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy. | Stars: 119k | Updated | Claude CodeCodex | Verified | Install |
| childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |