forge-checker-handoff
Governs the checker-handoff gate for FORGE verdicts that authorize an irreversible action. Use at the end of the Emit phase, after the verdict and recommended action are drafted, to check the recommended action against the five mechanically observable irreversibility trigger markers (deploy/release, destructive migration or data deletion, security-boundary change, external spend/commitment, public communication). On a match, set the emitted reasoning-report's requires_checker flag to true instead of letting the verdict flow straight to execution. Reversible verdicts (no marker match) are unaffected — do not fire this skill for them.
- 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
- forge-checker-handoff
- metadata
- { "phase": "E", "methodology": "FORGE" }
- description
- Governs the checker-handoff gate for FORGE verdicts that authorize an irreversible action. Use at the end of the Emit phase, after the verdict and recommended action are drafted, to check the recommended action against the five mechanically observable irreversibility trigger markers (deploy/release, destructive migration or data deletion, security-boundary change, external spend/commitment, public communication). On a match, set the emitted reasoning-report's requires_checker flag to true instead of letting the verdict flow straight to execution. Reversible verdicts (no marker match) are unaffected — do not fire this skill for them.
03 SKILL.md 正文
Checker-Handoff Skill — FORGE (maker ≠ checker, additive)
Loaded at the tail of the Emit phase, after the verdict and its recommended
action are drafted but before the envelope is finalized. FORGE's own Gate
(skills/verification/SKILL.md) is a self-review pass — sound for most
verdicts, but not sufficient authority to greenlight an action that cannot be
undone. This skill is the mechanical trigger that routes those verdicts to a
distinct checker before any action is taken.
When to use
Load this skill once per deliberation, at Emit, immediately after the winning
hypothesis's recommended action is known. Check the recommended action
against the trigger table below. Fire the hop only on a match. Do not use for
verdicts whose recommended action is reversible at ordinary cost — those emit
exactly as before, with requires_checker: false (the default).
Irreversibility trigger table
Five categories, checked mechanically against the recommended action's text — a match on any category fires the hop:
| # | Category | Observable markers in the recommended action |
|---|---|---|
| 1 | Deploy / release | deploy, release, ship to production, go live, roll out, promote to prod |
| 2 | Destructive migration or data deletion | drop table, truncate, delete from, irreversible migration, purge data, destroy backup |
| 3 | Security-boundary change | change permissions, rotate credentials, modify IAM/ACL, disable auth, open firewall, grant/revoke access |
| 4 | External spend / commitment | sign contract, purchase, commit budget, vendor agreement, procurement, SLA commitment |
| 5 | Public communication | press release, public announcement, customer-facing communication, publish blog post, public API contract change |
These five categories are the closed trigger set for this skill — do not
extend them ad hoc. If a recommended action is genuinely irreversible but
does not fit any of the five, the standard Reversibility dimension in
skills/framing/SKILL.md Step 4 and the verdict's [RISK] markers already carry
that signal; this skill only formalizes the mechanically-observable subset.
Your hop
- Detect — After drafting the recommended action, scan it against the trigger table above. Zero matches → skip to step 4 (no hop).
- Mark — On a match, set
requires_checker: truein the emitted reasoning-report's frontmatter (schemas/reasoning-report-profile.v1.jsonrequires_checkerfield). Add a[CHECKER-REQUIRED]line to the verdict's Provenance section naming the fired category and a checker-class hint: VIGIL for evidence-based/technical claims (root-cause, feasibility, reproducibility — claims a second reasoning or investigation pass can adjudicate), human for judgment calls (organizational, strategic, or policy trade-offs no evidence pass resolves). The hint is guidance, not a dispatch — see Invariants below. - Emit unchanged otherwise — The hop does not add a reasoning pass, does
not trigger REFORGE, and does not block emission. FORGE still emits its
normal verdict on the normal 1–3 pass / 1-REFORGE budget;
requires_checkeris a payload marker, not a refusal. - Default — Absent a trigger match,
requires_checkerisfalse(the schema default). Reversible verdicts are byte-for-byte unaffected by this skill's existence.
Invariants
- maker(forge) ≠ checker(vigil/human). FORGE's own Gate is necessary but
not sufficient authority for an irreversible action. A
requires_checker: truereport MUST be reviewed by a distinct party before any action — FORGE never self-certifies past this line. - Tool-less, always. FORGE holds no tools (P0-1) and does not gain any
at this hop. The hop is expressed purely via the emitted-artifact
marker (
requires_checker: true+ the[CHECKER-REQUIRED]note) — never a tool call, never a direct dispatch to VIGIL or a human. Routing between VIGIL and human is the orchestrator's decision, informed by but not bound to FORGE's checker-class hint. - Additive, never blocking. This hop composes with — does not extend — the existing 3-pass / 1-REFORGE cap and does not gate emission. A verdict that fails to emit because of this skill is a bug in this skill, not correct behavior.
- Backward compatible. Orchestrators that do not read
requires_checkersee no change in behavior — the field defaults tofalseand the rest of the reasoning-report is unchanged.
Graceful skip
If the recommended action's text is ambiguous or the decision type makes the
five categories inapplicable (e.g. a pure ROOT-CAUSE verdict with no
recommended action beyond "investigate further"), requires_checker stays
false. This skill never fabricates an irreversibility claim to be safe —
false positives erode the signal as much as false negatives.
Reasoner — Checker-Handoff Skill (maker ≠ checker, additive)
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.