kupo-keep-or-kick
Phase K triage gate that decides whether Kupo will attempt a delegated micro-task or refuse cheaply. Use when a delegation arrives and Kupo must decide KEEP vs REFUSE/ESCALATE; runs once per delegation at the start of every KUPO cycle. Do NOT use during Phase U/P/O.
- License 许可
- Apache-2.0
- 所属仓库
- Rynaro/magicite
- 最近更新
- 2026-10-14
- Stars
- ★ 2
- Forks
- 0
01 前言 Front Matter
- name
- kupo-keep-or-kick
- metadata
- { "phase": "K", "methodology": "Kupo" }
- description
- Phase K triage gate that decides whether Kupo will attempt a delegated micro-task or refuse cheaply. Use when a delegation arrives and Kupo must decide KEEP vs REFUSE/ESCALATE; runs once per delegation at the start of every KUPO cycle. Do NOT use during Phase U/P/O.
03 SKILL.md 正文
Keep-or-Kick Skill — Kupo (Phase K triage)
When to use
Load during Phase K (triage). Runs once per delegation. Triage cost ≈ 1 step. Do not load during Phase U, P, or O.
Loaded during Phase K. Runs once per delegation. Triage cost ≈ 1 step.
Phase K is the additive-proof gate: Kupo only spends cycles on tasks it can plausibly win. Every task that fails triage bounces cheaply, making Kupo structurally non-negative to the delegating planner's session.
Decision Tree (run in order — first failure exits)
Step 1 — Localization check
Does the task require changes to ≤ 2 files that form one coherent change?
- Yes → continue to Step 2.
- No (cross-cutting, > 2 files, or multiple unrelated concerns) →
REFUSE{SCOPE_TOO_BROAD}orESCALATE{to: apivr}for a cross-cutting refactor, orESCALATE{to: vivi}for a loop-native coding campaign.
One coherent change means: changing file A causes file B to change as a direct mechanical consequence (e.g., rename a symbol + update one call site), not a coincidental co-location.
Step 2 — Named-verifier predicate (structural KEEP gate)
Can you name a concrete external verifier that will declare pass/fail on the outcome? The verifier must be one of: test runner, typecheck, linter, compiler, diff, schema validator.
"I think it looks right" is not a verifier. "The TypeScript compiler will exit 0" is a verifier.
- Yes, verifier named → record it as
verifierfor Phase O. Continue to Step 3. - No nameable verifier →
REFUSE{NO_EXTERNAL_VERIFIER}.
Step 3 — Scope-class match
Does the task fall into a KEEP class from the §3 taxonomy?
KEEP classes (pass → continue):
| Class | Key signal |
|---|---|
| rename / symbol-move with compiler confirm | git grep shows bounded occurrences |
| import / path fix | broken import path, test as verifier |
| lockfile / dep-pin bump | version string change, build as verifier |
| config-key edit versus schema | add / change one config key, schema validate |
| lint / format autofix apply | pre-computed lint fix, linter exit-0 |
| mechanical fixture update | fixture output changed, diff or test as verifier |
| one-line failing-assertion fix | obvious off-by-one or stale expected value |
| template boilerplate | fill a template slot, diff as verifier |
| bounded grep-replace | scoped string substitution, test suite as verifier |
REFUSE / ESCALATE classes (fail → exit):
| Class | Routing |
|---|---|
| Open-ended reasoning, design, or planning | REFUSE{NOT_EXECUTOR_ROLE} |
| Cross-cutting refactor (> 2 files) | REFUSE or ESCALATE{to: apivr} |
| Ambiguous spec or unclear target | REFUSE{AMBIGUOUS_SPEC} — clarify upstream |
| Loop-native coding campaign | ESCALATE{to: vivi} or ESCALATE{to: apivr} |
| No nameable external verifier | REFUSE{NO_EXTERNAL_VERIFIER} (Step 2 already catches) |
If none of the KEEP classes match and no REFUSE class applies, treat as
REFUSE{SCOPE_CLASS_UNKNOWN} and surface the ambiguity to the delegating parent.
Step 4 — Economic gate
Is the expected pass-rate > ~0.20? The threshold is approximate: haiku→opus cost ratio implies Kupo must succeed on roughly 1 in 5 delegations to break even.
Ask: is the task clearly mechanical (high confidence → proceed) or speculative (low confidence → refuse)?
- High confidence (pass-rate clearly > 0.20) →
KEEP{verifier: <named>}. - Borderline (unsure) → default
REFUSE{PASS_RATE_UNCERTAIN}and return the task to the delegating parent with a note. - Low confidence (pass-rate clearly ≤ 0.20) →
REFUSE{PASS_RATE_LOW}.
Output format
Emit exactly one of:
KEEP{verifier: "<verifier command or class>"}
REFUSE{code: "<REFUSE_CODE>", note: "<one sentence for the parent>"}
ESCALATE{to: "<vivi|apivr|spectra|orchestrator>", code: "<ESCALATE_CODE>", note: "<one sentence>"}
The output of Phase K is the verifier field carried forward into Phase O.
No other state from K enters Phase U; K is a one-shot gate.
Notes
- Triage cost = ~1 step. K must be cheap; if K itself is expensive, the additive-proof breaks. Do not invoke atlas-aci during K — gather happens in U.
- Structural, not verbal. KEEP iff a verifier can be named; escalation triggers are structural (files > 2, loop-native pattern) not introspective.
- Loop-native campaigns always ESCALATE. If the task description mentions "iterative", "implement feature X end-to-end", "write the whole module", or similar — it's a campaign, not a micro-task. Escalate immediately.
- REFUSE is cheap and correct. The parent can always re-delegate to a heavier Eidolon. A fast REFUSE is more valuable than a slow failure.
Keep-or-Kick Skill — Kupo Phase K triage
04 同仓库更多 Skills
wine-dxvk-cache-clear
Clear a stale DXVK shader cache that is causing graphical corruption in a Wine or Proton game. Use when: a game shows corrupted textures or shader flicker that started after a driver or DXVK version change.
steam-download-region-fix
Fix a Steam client stuck at very low download speed by changing its download region. Use when: Steam downloads are far below the local internet speed for every game, not just one.
proton-battleye-eac-toggle
Toggle BattlEye or Easy Anti-Cheat runtime support for a Steam Proton game. Use when: a multiplayer game refuses to launch with an anti-cheat error under Proton.
vivi-verify-incoming
Load when reading any upstream artefact handed off to Vivi that carries a sibling .envelope.json. BLOCKING per ECL §6.2.2: the orchestrator MUST have verified the envelope's SHA-256 (eidolons verify-envelope --block / eidolons run --verify) and recorded a verify_pass before dispatch. If no verify_pass exists for the message_id, or a verify_fail is present, REFUSE to process the payload and hand back to the orchestrator. Symmetric receiver gate — every Eidolon enforces it identically.
vivi-parallel-tracks
Load ONLY under TRANCE authorization when the Plan phase yields N independent, non-overlapping implementation tracks (disjoint file sets). Operationalizes the TRANCE G4 form — parallel multi-track implementation in mandatory git-worktree isolation, max 5 clean-context tracks, a per-track verifier cascade, a non-fungible per-track ≤3 reflection budget, explicit stop conditions, and a single-threaded merge/aggregation step. Absent TRANCE gating, Vivi runs exactly as today: single-track A→P→I→V→Δ/R.
vivi-methodology
Full Vivi cycle reference — A(nalyze) → P(lan) → I(mplement) → V(erify) → Δ(Delta)/R(eflect). Use when the task is a non-trivial feature implementation in a brownfield codebase and you need the complete methodology reference: complexity routing, evidence-grounded planning rules, test-anchoring requirements, and failure-escalation thresholds.