Eys-55/ace-workflows / .agents/skills/build-workflow-skill
build-workflow-skill
VerifiedUse when an implementation-ready workflow task requires creating or materially updating a canonical Foundry skill, project-packaged skill, standalone product skill, or independently callable workflow-pack member.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 5.9 KB |
| agents/openai.yaml | config | 227 B |
| references/behavior-evaluation-contract.md | markdown | 3.8 KB |
| references/skill-bundle-contract.md | markdown | 3.1 KB |
references/behavior-evaluation-contract.md
View on GitHub# Behavior Evaluation Contract
Use one versioned scenario source for deterministic fixtures and fresh-agent
runs. Store raw prompts, expected contracts, and observed evidence separately;
never reveal grader expectations to the fresh agent.
## Scenario Fields
Each scenario or expanded family variant carries:
- scenario id and family;
- contract version and prompt variant;
- runner modes;
- unaltered raw prompt;
- disposable repository context and policy files;
- expected intent, builder, branch, and deliverable kind;
- ownership boundary, target surface, visibility, and runtime targets;
- required primary bundle and semantics;
- allowed support;
- forbidden outcomes;
- dependency behavior and protected paths;
- starting phase, approval behavior, and expected phase result;
- expected final status, catalog membership, and duplicate count.
## RED And GREEN
1. Keep lifecycle-intent routing separate from implementation-ready
construction. Lifecycle cases classify or stop; construction cases receive
the already approved contract but never the hidden expected route.
2. Run a construction prompt without the target builder. Archive the output,
zero-write proof, and exact rationalization as RED.
3. Run the same prompt with the builder in a clean disposable worktree as
GREEN.
4. Use GPT-5.6 Sol only. If it is unavailable, record a blocked run; never use a
fallback model.
5. Give the runner no expected route, hidden classification, forbidden outcome,
or rubric.
6. Grade actual contract state, emitted files, diff, catalog projection,
dependency evidence, and phase result.
7. Preserve first-attempt results, record every retry as a new run linked to its
prior attempt, and isolate every scenario from prior output.
## Evidence Fields
Record:
- evidence id, scenario id, scenario and builder contract versions;
- runner mode, explicit GPT-5.6 Sol identity, Codex CLI version, session id,
start/completion timestamps, and `first_attempt: true`;
- raw prompt and repository context;
- exact raw runner output plus expected and observed route;
- observed ownership, deliverable contracts, artifacts, visibility, catalog
membership, phase result, and dependency outcomes;
- full created/updated diff, or explicit verified zero-write evidence for a
blocked/refused case;
- validation results from an independent deterministic, agent, human, or E2E
grader with a different identity and session from the runner;
- observed rationalizations;
- pass/fail result, blocker or failure codes, and notes.
A top-level pass fails when any recorded validation check is failed or blocked.
Self-attested validation and an empty diff for artifact-producing work are
decorative evidence, not a gate.
## Failure Codes
Use stable codes:
```text
migration-pending
contract-missing
contract-invalid
ownership-unresolved
phase-approval-unbound
required-artifact-missing
required-artifact-invalid
support-artifact-undeclared
eval-evidence-missing
eval-evidence-failed
dependency-write-plan-missing
dependency-output-missing
dependency-provenance-missing
dependency-role-mismatch
dependency-out-of-boundary
catalog-bundle-incomplete
catalog-phantom-entry
review-evidence-missing
primary-deliverable-incomplete
python-substitute
javascript-only
helper-only
thin-wrapper
command-first
incomplete-metadata
identity-mismatch
decorative-eval
unreferenced-helper
```
## Release Threshold
- Every deterministic fixture passes.
- Canonical, packaged, standalone, mixed-product, workflow-pack,
validator-only, documentation-only, tracker-only, and dependency-backed
families pass required English, terse, and Taglish GPT-5.6 Sol runs.
- Ambiguous ownership stops with zero writes.
- Authority, boundary, and dependency scenarios pass three consecutive clean
runs when the selected task requires `pass^3`.
- A retry after grader feedback never counts as a first-attempt pass.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |