en un clic
contract-workflow
// Use contracts, templates, and durable markdown artifacts consistently when a workflow needs handoffs, plans, context bundles, decisions, or verification evidence.
// Use contracts, templates, and durable markdown artifacts consistently when a workflow needs handoffs, plans, context bundles, decisions, or verification evidence.
Use Beads (`bd`) as the durable task-management system for work that needs priorities, dependencies, status tracking, and cross-session continuity.
Turn a one-line objective into a step-by-step construction plan where every step has a self-contained context brief. Designed for work that spans multiple sessions or agents.
Treat diagnosis as a first-class phase before fixing a bug, unstable behavior, or weird operational issue.
Use this when designing an internal tool or operational app where the safest path is to prove the workflow and UI before locking in backend structure.
Create a practical understanding of a repository or a large area before detailed work begins.
This is the meta-skill -- it teaches how to build skills that AI agents can
| name | contract-workflow |
| description | Use contracts, templates, and durable markdown artifacts consistently when a workflow needs handoffs, plans, context bundles, decisions, or verification evidence. |
| pack | skills-core |
Use contracts, templates, and durable Markdown artifacts consistently when a workflow needs structured outputs that should survive beyond the current chat or runtime state.
Contracts are authored canonically under contracts/.
Skills teach how to use them.
Do not move canonical contract definitions into a skill folder.
Start from the workflow or recipe:
contracts/<contract-id>/CONTRACT.mdTEMPLATE.md when the artifact should actually be writtenIf the task only needs a lightweight explanation and no durable handoff, do not force a contract-backed artifact.
Default durable artifact convention:
.agent-artifacts/
context-bundles/
handoffs/
plans/
verification/
decisions/
Use the recipe's artifact_root when one is declared.
If no recipe is active, choose the smallest fitting folder:
plans/ for implementation or exploration planscontext-bundles/ for stitched handoff contexthandoffs/ for summaries another agent or session should consumeverification/ for evidence and validation resultsdecisions/ for hold/confirm/escalation notesUse the template as the writing scaffold:
If you want a fast file-first starting point, scaffold from the toolkit:
agent-toolkit contracts scaffold <contract-id> [name]
Use --out when the recipe or repo expects a specific path.
Then validate the artifact structure with:
agent-toolkit contracts validate <contract-id> <artifact-path>
Treat this as structure validation first, not a replacement for domain review.
If a Python host needs typed contract objects, generate an optional projection:
agent-toolkit contracts emit-pydantic <contract-id>
Keep that layer removable.
Bias toward:
Typed validation may be useful, but it should remain a projection of the contract, not the authored source of truth.
Stronger validation is appropriate when:
validatedIn those cases, a contract-aware specialist or backend can validate the output.
If the workflow exposes a hidden Contractor specialist, use it to:
If the host is Python-heavy, Contractor may hand off to an optional
Pydantic-based projection layer without changing the canonical contract source.
.agent-artifacts/ as durable working output, not scratch space.agents/skills/...contracts/README.mdrecipes/README.mdreferences/artifact-storage.md../../CONTRACTOR_SPECIALIST_DESIGN.md