--- name: swe-retro description: Review an engineering session or repeated workflow friction and propose evidence-backed improvements to tools, checks, navigation, and guidance. Use for a requested retrospective; do not add a retrospective to every task. metadata: version: 1.0.1 updated: 2026-10-15 status: stable --- # Improve the engineering workflow ## Overview Find a small number of changes that would prevent demonstrated mistakes or reduce repeated effort. Read the applicable communication, general engineering, and language styles selected by active instructions and their relevant references. Preserve confirmed preferences; an incident does not establish a new personal rule. Use the current session unless the user names another scope. A retrospective is useful after recurring friction or significant work, not as a mandatory closing ceremony. No findings is an acceptable result. ## Steps 1. **Review the evidence.** Inspect the scoped conversation, changes, checks, and relevant repository configuration. Use a specified session artifact when available; avoid broad searches of unrelated private conversations. Identify what actually went wrong or consumed effort and what already worked. Separate observed failures from plausible causes and unsupported interpretations. 2. **Find the smallest useful improvement.** Consider these only where the evidence points to them: - Navigation: a missing ownership or documentation pointer caused repeated searching. Fix the pointer rather than duplicating the source content. - Automated checks: inspect existing lint, type, test, and CI commands first. Repair a disconnected or misleading check before introducing another one. A repeatable mechanical mistake may warrant a deterministic check when its prevention value exceeds its maintenance cost. - Guidance: clarify an ambiguous rule, remove a stale or duplicate instruction, or route specialized material only when relevant. Keep personal preferences in the owning private skill, not in a public workflow or copied into every project. Standards guide implementation and review. - Tool and information access: reduce redundant calls, narrow noisy output, or make needed diagnostics available. Distinguish a missing capability from a tool that was available but used poorly. - Workflow: remove unnecessary questions, artifacts, repeated checks, or orchestration that did not improve the result. Preserve checks required by the repository or justified by new evidence. 3. **Assess the trade-off.** For each supported candidate, state the observed incident, proposed correction, expected benefit, maintenance cost, and how its effect could be checked. Prioritize recurring or consequential failures. Prefer deletion or a local correction when it solves the problem. Avoid creating a new skill, policy, or system for an isolated reversible mistake. Recommend an experiment when the cause remains uncertain. 4. **Deliver within scope.** Report the highest-value actionable candidates concisely, using the owner's terminology and evidence. Do not pad the report to meet a finding count or create a new document solely to record the ritual. A request for retrospective findings alone does not authorize changes to configuration, tooling, or active personal rules. If implementation is already requested, apply authorized corrections and validate them with relevant checks. A new inferred preference remains a proposal until the owner confirms it; explicit decisions already made need no repeated confirmation. Preserve unrelated work and follow repository commit requirements. Report what changed and what remains a proposal. ## Completion checks - Findings are tied to evidence in the requested scope and distinguish causes from hypotheses. - Existing tooling and authoritative guidance were checked before recommending additions. - Recommendations explain their benefit, cost, and smallest practical next step. - Proposed rules are distinguished from confirmed preferences, and public workflows contain no copied personal guidance or private source evidence. - Applied changes were authorized and verified; proposals and limitations remain explicit. No extra process was added merely to produce a retrospective.