khujta Gabe Scope Change — classify, then evolve
Usage: /gabe-scope-change [--force-addition | --force-pivot] <description>
Gabe execution contract (E1–E7)
This skill runs under the suite execution contract — E1 EVIDENCE · E2 RUN-BEFORE-✅ · E3 NO SILENT DOWNGRADE · E4 REUSE FIRST · E5 STATE SYNC · E6 MISSING ANCHOR = STOP · E7 REPORT WHERE — floors, not ceilings; a skill's own gate may be stricter, never looser. Full text: ../gabe-docs/references/execution-contract.md (if that file is missing, E6 applies — STOP).
Single entry point for modifying a finalized SCOPE.md — its premise sections or its ## Phases section alike. Classifies the requested change; additions are executed by this skill's own Addition path (Step 5); pivots route to /gabe-scope-pivot, which stays a separate skill on purpose — its disable-model-invocation flag structurally prevents a model from self-triggering a direction change, and a merged mode cannot carry that flag.
Why classifier-first and not user-picked? Misclassifying a pivot as an addition corrupts version history. The classifier uses the declared 9 rules + Opus reasoning + rationale, so the user sees why. Overrides exist, but the default prevents silent scope corruption.
Rendering note. Output templates wrapped in bare triple-backtick fences are spec-meta delimiters — render as plain markdown at runtime. Tagged fences (```jsonl) stay fenced. See
gabe-docs/SKILL.md§ "Runtime output rendering convention".
Procedure
Step 1: Pre-flight
- If
.kdbp/SCOPE.mddoes not exist: exit with "No finalized SCOPE.md. Run/gabe-scopefirst." - If
.kdbp/scope-session.jsonexists (in-progress scope): exit with "Active scope session in progress. Finish or abort/gabe-scopefirst." - Parse
$ARGUMENTS: extract--force-addition,--force-pivot,--force(coverage bypass, Addition path only), and description text. - If no description provided: prompt "What do you want to change? (one paragraph describing the desired change)"
- Read SCOPE.md (incl.
## Phases) +scope-references.yamlinto structured state.
Step 2: Classify
If --force-addition or --force-pivot set → skip classification, go to the matching path.
Otherwise invoke prompts/scope-change-classifier.md (Opus) with {current_scope, proposed_change, user_intent}.
The 9 pivot triggers (any otherwise addition) — copied verbatim from ~/.claude/prompts/gabe-scope/scope-change-classifier.md §System role, keep in sync:
| # | trigger_rule | Fires when |
|---|---|---|
| 1 | primary_user | Primary User changes (role/persona/segment) |
| 2 | non_user_flip | a Non-User becomes a Primary/Secondary User, or vice versa |
| 3 | sc_change | a Success Criterion is removed, inverted, or truth-flipped |
| 4 | goal_flip | a Non-Goal becomes a Goal, or a Goal becomes a Non-Goal |
| 5 | posture_shift | Architecture Posture macro-shift (sync↔async, local↔cloud-first…) |
| 6 | ref_conflict | authoritative ref replaced/downgraded/overridden, or a new conflicting authoritative ref added |
| 7 | business_model | funding/business-model shift that retargets the product |
| 8 | constraint_infeasibility | a constraint change makes an existing SC/REQ infeasible as written |
| 9 | timeline_compression | timeline forces REMOVING/skipping phases, not just accelerating |
Fallback gate: if the classifier prompt file cannot be read, do NOT classify from memory — present this table to the user and ask them to pick, recording classifier: unavailable — user-picked rule <N> in CHANGES.jsonl.
Output: {classification, trigger_rule, rationale, confidence, user_intent_matches_classification, suggested_next_command}.
Step 3: Present + confirm
Render the classification (classification · trigger · confidence · rationale · suggested path). Options:
[p]Proceed with the suggested path[o]Override to the other path (requires a one-line rationale, appended to the Change Log)[c]Cancel — exit, no writes
Low-confidence classifications add: "Classifier confidence low — human review recommended before proceeding." Record every routing decision in .kdbp/CHANGES.jsonl:
{"ts":"2026-04-21T15:00:00Z","event":"scope_change_classified","classification":"pivot","trigger_rule":"primary_user","confidence":"high","override":false,"routed_to":"/gabe-scope-pivot"}
Step 4: Pivot path — route out
Classification = pivot → exec /gabe-scope-pivot with the description as its argument; this skill exits. All pivot writes happen there.
Step 5: Addition path — execute inline
Classification = addition → this skill performs it. Invariants:
- SCOPE.md
scope_version:stays the same (additions never bump major);last_scope_event:→ today;phases_version:bumps by 1. - Change Log (§15) appends an
additionrow — never rewrites existing rows. - Prior phase IDs and REQ IDs are immutable. New phases use decimal IDs; new REQs append with the next sequential ID.
5a — Detect addition type (ambiguous → prompt; never guess):
| Type | Example | Action |
|---|---|---|
| New REQ | "Add REQ for export-to-markdown" | Append REQ-NN+1 to §12; map to existing SC or new phase |
| New SC (non-inverting) | "Add SC tightening REQ-02's acceptance signal" | Append SC-NN+1 to §7; wire to existing REQ |
| New phase | "Insert observability phase between 2 and 3" | Insert with decimal ID 2.1 |
| New reference | "Add suggestive ref for wireframes" | Append to scope-references.yaml (contextual/suggestive only; authoritative adds are trigger 6 → pivot) |
| Constraint tightening | "Tighten budget to $10/mo" | Update §9; must stay feasible for existing SCs (else trigger 8 → pivot) |
| Non-goal addition | "Add NG for multi-tenant" | Append NG-NN+1 to §8 |
5b — Draft with Sonnet via the matching prompts/ drafter (users-and-non-users / non-goals-generator / req-decomposer …), passing current SCOPE state + the description.
5c — Write targets (all appends, never rewrites): .kdbp/SCOPE.md (new entity + anchor); its ## Phases (decimal-ID insert + Phase Table + Dependency Graph when adding a phase); scope-references.yaml (new ref); SCOPE §15 Change Log (addition row — one log for premise and phase-arc changes alike).
5d — Coverage re-check (same deterministic validator as /gabe-scope Step 7): every SC has ≥1 REQ; every REQ in exactly one phase. Broken coverage → prompt "Coverage now incomplete. Add REQ for new SC now? [y/N]" and loop until fixed or --force (records finalize_forced_at + rationale).
5e — Confirm + commit. Diff preview → user approves → write files → offer the git-commit prompt (same flow as /gabe-scope Step 8f). Append to CHANGES.jsonl:
{"ts":"2026-04-21T15:30:00Z","event":"scope_addition","scope_version":1,"phases_version":2,"addition_type":"new_phase","entity_ids":["2.1"]}
Decimal ID rules: between integers N and N+1 → N.1, N.2, …; after N.1 → N.2; no nested decimals (N.1.1 forbidden — if depth is needed, that's a pivot); existing integer IDs never renumber; collisions auto-assign the next free decimal.
Flags
--force-addition/--force-pivot— skip the classifier (recordsoverride: true; a one-line reason is still required in the Change Log). Both set → reject: "Pick one override flag."--force— Addition path only: bypass the coverage re-check.
Edge cases
- Addition that implicitly becomes a pivot (e.g., "add REQ" whose text changes an SC's observable truth): a mini-classifier call on the drafted output detects it → abort with "This addition appears to modify existing SC semantics. Pivot recommended — re-run, or
--force-additionto override." - Description touches both pivot AND addition scope: the classifier picks the more disruptive (pivot); the rationale surfaces both; user can split into two invocations.
- Empty description twice → exit.
Integration
Called by the user directly; does not chain from /gabe-scope. Pivot writes happen in /gabe-scope-pivot; addition writes happen here. (The former standalone /gabe-scope-addition entry point is archived — skills/_archive/ — its procedure lives on as Step 5.)