Modify wrap-up skill to recommend immediate action on deferred items when appropriate
Problem
The wrap-up skill treats all deferred findings uniformly — every item becomes a "create ticket" action. Many findings (unused imports, missing return types, typo fixes) are trivial enough that creating a ticket and fixing them in a separate branch is more overhead than the fix itself. There's also no prompt to create a PR after wrap-up, leaving a gap in the session-closing workflow.
Context
Multiple skills already make overlapping "is this simple enough to just do?" judgments using ad-hoc prose criteria:
next-steps-after-plan rule 2: "mechanical and isolated"next-steps-after-review: "simple and well-understood"orchestrated-architect: none/low impact classifications
These should share a single rubric rather than reinventing the criteria independently.
Solution
1. Create _data/complexity-classification.md — a shared reference doc defining a four-level complexity rubric (trivial, mechanical, involved, architectural) with clear characteristics for each level. Each consuming skill defines its own threshold against this rubric.
2. Add a quick-fix pass to wrap-up — after building the inventory (Phase 1), classify each finding's complexity. Items at levels 1–2 (trivial/mechanical) are presented as quick-fix candidates in a dedicated pass before the standard housekeeping menu. Applied items are committed and removed from the findings pool. Skipped items demote into the standard "create ticket" flow.
3. Add a PR prompt to wrap-up — after all wrap-up actions complete, if there are code changes on the branch, prompt the user to create a PR via /summarize-change.
4. Update existing consumers — replace inline complexity criteria in next-steps-after-plan and next-steps-after-review with references to the shared rubric.
Acceptance criteria
Modify wrap-up skill to recommend immediate action on deferred items when appropriate
Problem
The wrap-up skill treats all deferred findings uniformly — every item becomes a "create ticket" action. Many findings (unused imports, missing return types, typo fixes) are trivial enough that creating a ticket and fixing them in a separate branch is more overhead than the fix itself. There's also no prompt to create a PR after wrap-up, leaving a gap in the session-closing workflow.
Context
Multiple skills already make overlapping "is this simple enough to just do?" judgments using ad-hoc prose criteria:
next-steps-after-planrule 2: "mechanical and isolated"next-steps-after-review: "simple and well-understood"orchestrated-architect:none/lowimpact classificationsThese should share a single rubric rather than reinventing the criteria independently.
Solution
1. Create
_data/complexity-classification.md— a shared reference doc defining a four-level complexity rubric (trivial, mechanical, involved, architectural) with clear characteristics for each level. Each consuming skill defines its own threshold against this rubric.2. Add a quick-fix pass to wrap-up — after building the inventory (Phase 1), classify each finding's complexity. Items at levels 1–2 (trivial/mechanical) are presented as quick-fix candidates in a dedicated pass before the standard housekeeping menu. Applied items are committed and removed from the findings pool. Skipped items demote into the standard "create ticket" flow.
3. Add a PR prompt to wrap-up — after all wrap-up actions complete, if there are code changes on the branch, prompt the user to create a PR via
/summarize-change.4. Update existing consumers — replace inline complexity criteria in
next-steps-after-planandnext-steps-after-reviewwith references to the shared rubric.Acceptance criteria
_data/complexity-classification.mdexists with four levels, clear characteristics, and consumer-threshold guidancenext-steps-after-planrule 2 references the shared rubric instead of inline criterianext-steps-after-review"simple and well-understood" routing references the shared rubric