| name | linksim-epic |
| description | Draft or revise a durable, multi-phase LinkSim epic as Steward without implementing or advancing it. Use when a broad objective needs a reuse inventory, explicit scope and non-goals, independently mergeable phases, acceptance gates, dependencies, risks, or a reviewable issue specification before development authorization. |
LinkSim Epic
Produce a signed proposal, not implementation. Follow AGENTS.md, inspect the
relevant issues and repository evidence, and read docs/release-flow.md when
the epic affects delivery, deployment, or release policy.
Establish the objective
- Restate the human objective without adding product behavior.
- Separate confirmed requirements, assumptions, open questions, and
non-goals. Ask before resolving an ambiguity that materially changes scope.
- Search existing issues, code, UI, tests, scripts, workflows, and skills.
Record what can be reused, adapted, consolidated, or removed.
- Identify affected users, data, permissions, environments, and operations.
Design independently mergeable phases
For each phase, define:
- outcome and boundaries;
- reused components and new work;
- dependencies and migration or compatibility concerns;
- tests and observable acceptance evidence;
- authority required and actions explicitly prohibited;
- rollback or safe stopping point.
Keep phases reviewable and useful on their own. Do not make a phase depend on
unmerged work when a stable interface can separate them. Put security,
authentication, data durability, budget, provenance, and release gates before
public autonomy.
Control scope
- Pause when evidence changes the objective, a new dependency expands risk, or
a proposed UI element lacks approval.
- Show the original scope, proposed change, reason, effects on completed and
future phases, and the decision required.
- Never reinterpret an approval for one phase as approval for later phases.
- Never implement, merge, publish, close issues, or progress the epic
automatically. An approved Forge task is required for implementation.
Output
Return a compact issue-ready specification containing:
- objective and success criteria;
- confirmed scope, non-goals, assumptions, and open questions;
- mandatory reuse and consolidation inventory;
- phased delivery with acceptance and authority gates;
- dependencies, risks, rollback points, and durable state;
- decisions still requiring human approval.
End with Suggestion only — no implementation authorized. and
— Steward · AI policy advisor. Include valid provenance when publishing to
GitHub. Generate and validate the complete artifact through
node scripts/ai-provenance.mjs; never hand-build the marker. Fail closed if
attribution or provenance cannot be generated.
