issue_80: Present the toolkit through capabilities - #83
Conversation
Review: change needed before mergeThe capability map is well structured, and the distinction by entry point makes the overlap between Spec-Driven Development and Decision/Impact Management understandable. The validation workflow is also green. One acceptance criterion of #80 is nevertheless knowingly left unmet: the Spec-Driven Development capability must describe the workflow “from feature request to convergence”, while this version ends at Reviewable Architecture Delivery. Because the PR also uses Please resolve this before merge. The cleanest sequence seems to be:
Alternatively, keep this PR independent but do not close #80 until a concrete follow-up in the same delivery chain completes that criterion. Given that #84 already provides the missing canonical target, wiring it in here after #84 lands is clearer and avoids advertising a nonexistent skill. |
A flat skill list explains the building blocks and hides the strongest claim: eight skills together already deliver an architecture-aware Spec-Driven Development workflow, and nothing in the README said so. The new section sits above everything else, because presentation was the problem. Each capability names the moment it starts from, the outcome, and the skills that collaborate — as navigation, never as a second copy of the rules. Two findings from validating the proposed map against the current skills: - Decision and Impact Management is a strict subset of Spec-Driven Development by membership. It is a real capability only when framed by entry point: a feature request versus an architecture question arriving on its own. Framed by skill list it would teach a distinction that does not exist. - Session Continuity is a group of its own. clock-in and clock-out are the only skills a user invokes by name, daily, and a feature view that omits them describes the toolkit as less usable than it is. presentation is named outside the map rather than stretched into a group it does not belong to. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01USXp58FoRppFK6FUht8K7u
Review on #83: the Spec-Driven Development entry stopped at Reviewable Architecture Delivery while #80 requires the workflow to be described "from feature request to convergence", and the PR closes that slice. The criterion would have been closed unmet. The gate exists now that #81 has landed, so the entry says what the capability actually produces: a change whose request, specification, architecture knowledge, implementation, tests and delivery metadata tell one consistent story — checked rather than assumed, with the four result states named and the non-repair rule stated. Reviewable Architecture Delivery gains the same gate before a pull request is declared mergeable, which is the second place the check applies. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01USXp58FoRppFK6FUht8K7u
|
Reworked in 529d61c, following your prescribed sequence: #84 merged, this rebased onto it, You were right that the alternative — keep this independent and leave #80 open — was the worse option. Spec-Driven Development now states what the capability actually produces rather than where its skills stop:
and closes with the gate, its four result states, and the non-repair rule. Reviewable Architecture Delivery gained the same gate before a pull request is declared mergeable. That is the second place the check applies, and leaving it out would have implied the gate belongs only to feature work. Verification, all mechanical:
All acceptance criteria of #80 are now met, including the one this PR previously deferred. One consequence for #82: its first carry-over is done. The remaining work there is the wiring into |
Re-review: implementation is sound; PR description still needs alignmentThe rework in 529d61c resolves the substantive finding:
One delivery-metadata inconsistency remains in the PR body:
Please update or remove those stale passages. This is especially worth fixing here because the new gate explicitly asks whether the pull request and source documentation describe the same resulting state. Once the PR description is aligned, I see no remaining blocker to merge. |
|
Description aligned: skill count corrected to 20, and the "deliberately incomplete" section removed — both statements it made were obsolete after #84 and 529d61c. Worth recording why this one stings: the stale passages were a question-7 finding against this very pull request — do the issue, the pull request, the source documentation and the code describe the same resulting state? They did not. The code converged and the delivery metadata did not follow, which is the exact drift the gate exists to catch, found by a human on the pull request that introduces it. I have added a short note at the end of the description saying so, rather than quietly editing the numbers. The first use of a gate is worth keeping visible. Merging now, then continuing with #82. |
Summary
Adds a Capabilities section to
README.md, above everything else, because presentation was the problem it exists to fix. Six capabilities, each naming the moment it starts from, the outcome it produces, and the canonical skills that collaborate to produce it.The strongest claim the toolkit can make was previously invisible: eight skills together already deliver an architecture-aware Spec-Driven Development workflow, from feature request through to a checked convergence, and a flat list of eighteen bullets cannot show that.
Closes #80
Feature map validated before writing
The first acceptance criterion, recorded on the issue. Two findings changed the map #77 proposed:
Decision and Impact Management is a strict subset of Spec-Driven Development — all five of its skills also appear there. #77 permits a skill in several capabilities, which is right, but a feature that is a subset of another feature is a phase, not a peer. It becomes a real distinction when framed by entry point: a feature request versus an architecture question arriving on its own. Both entries are now written that way, and the overlap with Reviewable Delivery is stated rather than hidden.
Session Continuity is a group of its own.
clock-inandclock-outpost-date #77 and appear in none of its five groups. They are also the only skills a user invokes by name, daily — omitting them describes the toolkit as less usable than it is.Five skills the proposed map left unplaced are now placed:
commit-messageunder Reviewable Delivery,clock-in/clock-outunder Session Continuity,architecture-coreas a precondition above the list rather than a member of one, andpresentationnamed explicitly outside the map. A feature map that stretches to include everything stops describing what the toolkit is for.The arc closes through the Convergence Check
Rebased onto #84 after review. Spec-Driven Development no longer stops where its skills stop; it states what the capability produces:
and closes with the gate, its four result states, and the non-repair rule. Reviewable Architecture Delivery uses the same gate before a pull request is declared mergeable — the second place it applies, and leaving it out would have implied the gate belongs only to feature work.
Verification
convergence-checkincluded — checked mechanically againstskills/**/SKILL.mdfrontmatter.README.mdresolves to an existing path — checked mechanically../build.sh test— 35 JS tests, 157 Ruby assertions, 0 failures../build.sh check-adapters— current.Every acceptance criterion of #80 is met, including "from feature request to convergence", which an earlier revision of this PR deferred while the gate did not yet exist.
Notes for the reviewer
Placement is the deliberate part: the section goes before "Relationship to docToolchain", not next to the skill list. A capability overview buried at line 245 would not have solved the problem that produced #77. The existing "AI Skills" list stays exactly as it was and is now introduced as the reference view, with a link back up.
The stale passages an earlier version of this description carried — a skill count of 19, and a section claiming the Convergence Check did not exist — were exactly what question 7 of the new gate asks about: whether the issue, the pull request and the source documentation describe the same resulting state. They did not. Fixed.