AGENT SKILL

gilgamesh-attest

Phase A (Attest) result emission — finalizes the TaskState audit record with evidence-anchored claims, emits the mission result via ECL PROPOSE, and PROPOSEs any typed handoff-request UPWARD (never dispatches). Use when Grind reaches a terminal stop_state; load at the end of a GILGAMESH cycle. Do NOT use during Gauge/Inventory/Lock/Grind.

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
gilgamesh-attest
metadata
{ "phase": "A", "methodology": "Gilgamesh" }
description
Phase A (Attest) result emission — finalizes the TaskState audit record with evidence-anchored claims, emits the mission result via ECL PROPOSE, and PROPOSEs any typed handoff-request UPWARD (never dispatches). Use when Grind reaches a terminal stop_state; load at the end of a GILGAMESH cycle. Do NOT use during Gauge/Inventory/Lock/Grind.

03 SKILL.md 正文

Attest Skill — Gilgamesh (Phase A result)

When to use

Load when Grind reaches a terminal stop_state (terminate or escalate). Attest turns the mission into an audit record and an ECL PROPOSE. Never invoked in Gauge, Inventory, Lock, or Grind.

The wanderer leaves with no weapons: everything collected during the mission is either returned here in the Attest record or discarded. Nothing persists (security.persists: []).


Step 1 — Finalize the TaskState (the audit record)

Complete schemas/taskstate.v1.json. Each entry in claims[] MUST carry ≥1 external evidence anchor drawn from verification_outcomes[] (an oracle result, a path:line, or a diff). A claim with no external anchor is inadmissible — the Excalipoor rule: a proud assertion is worthless if no oracle confirmed the blade.

Record: final stop_state, verification_outcomes, changed_artifacts (sandbox paths), unresolved_risks, and the spent budget. This is the mission's single audit artefact; ship it verbatim.


Step 1a — The report is the attestation (mission report protocol)

When the inbound mission enumerates required labeled report lines, your final human-readable message is itself part of the audit record, not a separate courtesy summary. A report missing any required line — or one that skips enumeration — is an inadmissible attestation, the same Excalipoor rule an anchor-less claim fails.

1. Enumerate first. The FIRST line of the final report is REQUIRED-LABELS: followed by every required label copied from the mission. RE-READ the mission and build this line before drafting any answer — a report that skips enumeration is invalid regardless of what follows. Do this even under budget pressure.

2. Label discipline. Then one line per label, LABEL: value. Reproduce each label exactly as the mission gives it, VERBATIM but WITHOUT any parenthetical hint: in EVIDENCE-x (path:line) the (path:line) describes the value's shape and is never part of the label. Write EVIDENCE-x: <path>:<line>. Folding the hint into the label turns a matched line into an unmatched one under mechanical grading.

3. Value format. The value's FIRST whitespace-delimited token IS the answer — a number, pass/fail, or a path:line — with explanatory annotation trailing after a space. A grader reading the first token should never have to parse prose to find the answer.

4. Never omit a line. If a verification is blocked, its line is still emitted with value fail plus the blocker as annotation — never a missing line, and never a pass you did not actually observe.

Quoted-anchor rule (Excalipoor applied to citations). Every path:line anchor you emit must be followed by a space and a short double-quoted verbatim fragment (3–6 words) of that exact line — copied AFTER Reading the line, never from memory. If you cannot quote it, you have not read it; do not cite it. Example: EVIDENCE-enum_count: schemas/roster-entry.schema.json:11 "scout, planner, coder". Prefer a single line over a range.

Pre-emit anchor re-read rule (mechanical, not exhortative). Before writing ANY path:line anchor into the final report, Read that exact line number in that file and confirm the line's text contains the verbatim fragment being quoted. If the line does not contain the fragment (or is empty/out of range), the anchor is WRONG — re-locate the true line with Grep on the fragment, and cite the corrected line number. Never emit an anchor you have not re-read at that exact line in this same mission; the quoted fragment must be present on the cited line. A drifted anchor is an inadmissible attestation — the Excalipoor rule applied to line numbers: a citation you cannot re-read is a fake blade. This closes the gap a quoted fragment alone does not: a fragment copied once at write-time can still drift onto the wrong line by the time the report is assembled — the re-read happens immediately before the anchor is emitted, not earlier in the mission.

Repo-only anchor rule. Evidence and proposal-target anchors MUST point at a repo-relative path that exists in the working checkout (a committed source file). NEVER cite an ephemeral, temp, sandbox-scratch, or command-output path (/tmp/*, redirected stdout, etc.). If a fact is produced by RUNNING a command, the command's result belongs in the VERIFY-<name> line; the EVIDENCE-<key> anchor points at the repo file or script the fact derives from (e.g. cite the script's source line, not its runtime output file).

Verify-routing ladder. (1) If the verification command is directly allowed by your tool list (make, bats, shellcheck, pytest, go test, jest, rspec, shasum, wc) — run it directly. (2) Otherwise route it through the delegated sandbox: eidolons sandbox run --allow-unsafe-host -- <cmd> (see skills/grind/SKILL.md — externalized verification; the sandbox captures pass/fail). (3) Only if both rungs are unavailable, emit VERIFY-<x>: fail immediately followed by the blocker. Never skip a rung, never omit the line, never hand-derive a result you could execute.


Step 2 — Emit the result (ECL PROPOSE)

On terminate (success), emit the mission result to the orchestrator:

  1. Write the result artefact + compute its sha256 and size_bytes.

  2. Compose the ECL v2.0 envelope sidecar (validate against schemas/ecl-envelope.v2.json):

    {
      "envelope_version": "2.0",
      "message_id": "<UUIDv7>",
      "thread_id": "<from inbound envelope>",
      "from": { "eidolon": "gilgamesh", "version": "<version>" },
      "to": { "eidolon": "orchestrator" },
      "performative": "PROPOSE",
      "objective": "<=240-char summary of the result>",
      "artifact": { "kind": "<result-kind>", "schema_version": "1", "path": "<rel>", "sha256": "<hex>", "size_bytes": <N> },
      "ise": {
        "assertion_grade": "validated",
        "provenance": { "methodology_version": "gilgamesh-<version>", "tool_surface": ["Read","Grep","Glob","Bash(eidolons sandbox:*)"], "lateral_consults": [] },
        "receiver_authorization": { "auto_route": true, "auto_merge": false, "auto_deploy": false }
      },
      "integrity": { "method": "sha256", "value": "<hex>" }
    }
    

    assertion_grade: "validated" is earned by construction (§4 pre-completion gate), never self-report. auto_merge: false is load-bearing: the parent applies and commits (PROPOSE-only). auto_deploy: false mirrors the never-grantable deploy row. lateral_consults records a forge consult if one informed the result; otherwise empty.

On escalate, emit ESCALATE instead, carrying the last oracle output.


Step 3 — Emit handoff-request(s) UPWARD (never dispatch)

For any work that crossed an information / authority / modality boundary, emit a typed handoff-request (schemas/handoff-request.v1.json) as a PROPOSE to the orchestrator — a proposal to route, not a dispatch (AC-E09, HC8):

{
  "schema_version": "1",
  "kind": "handoff-request",
  "objective": "<sub-mission goal>",
  "scope": { "paths": ["..."], "mode": "read-write" },
  "deliverables": ["..."],
  "evidence_required": ["<oracle output that will prove it>"],
  "stop_conditions": ["..."],
  "suggested_specialist": "<atlas|kupo|vigil|idg|forge|...>",
  "evidence_of_boundary": {
    "boundary_kind": "authority",
    "rationale": "requires deploy authority Gilgamesh's table denies (never-grantable)"
  }
}

evidence_of_boundary is REQUIRED. Delegation is never justified by "it looks complex" (digest §5) — it must name the boundary: a separable information context, an authority Gilgamesh cannot grant itself, or a distinct modality Gilgamesh cannot verify externally.

Cross-check the worker-never-router invariant: handoffs.downstream: []. Gilgamesh invokes no spawn primitive and calls no subagent. The orchestrator owns routing; Gilgamesh only proposes it.


Step 4 — Memory + session end (graceful skip)

After emitting PROPOSE:

  • mcp__crystalium__ingest(envelope, payload) — record the outbound spine for provenance. There is no post-flight commit: unlike a specialist, Gilgamesh grows no durable pattern library (R-032).
  • mcp__crystalium__session_end() on the terminal exit → Dream consolidation.

All mcp__crystalium__* calls are skipped silently when CRYSTALIUM is absent.


Notes

  • Evidence or it did not happen. Every claim resolves to an oracle result. No anchor → drop the claim; it was never real.
  • Result PROPOSE, never commit. The parent applies and commits with its own authority; Gilgamesh holds none.
  • Handoff is a proposal, not a dispatch. You return the typed request; the orchestrator routes it. You never spawn.

Attest Skill — Gilgamesh Phase A: audit record + evidence-anchored PROPOSE + upward handoff-request

04 同仓库更多 Skills

Rynaro
Rynaro/magicite

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.

★ 2
Rynaro
Rynaro/magicite

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.

★ 2
Rynaro
Rynaro/magicite

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.

★ 2
Rynaro
Rynaro/magicite

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.

★ 2
Rynaro
Rynaro/magicite

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.

★ 2
Rynaro
Rynaro/magicite

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.

★ 2