idg-verification
Defines the CHT verification framework and provenance output for the Gate phase of the IDG cycle. Use when entering the Gate phase, verifying document completeness and truthfulness, running CHT scores, deciding DELIVER vs REVISE vs ESCALATE, or producing provenance metadata blocks.
- 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
- idg-verification
- metadata
- { "phase": "G", "methodology": "IDG" }
- description
- Defines the CHT verification framework and provenance output for the Gate phase of the IDG cycle. Use when entering the Gate phase, verifying document completeness and truthfulness, running CHT scores, deciding DELIVER vs REVISE vs ESCALATE, or producing provenance metadata blocks.
03 SKILL.md 正文
Verification Gates
Loaded during the Gate phase. Defines the CHT verification framework and provenance output.
When to use
Load this skill when entering the Gate (G) phase of the IDG cycle — specifically when running the CHT gate (Completeness / Helpfulness / Truthfulness), generating provenance metadata, or deciding whether to deliver, revise, or escalate. Also load for the section-level mini-gate in G5 parallel mode. Do NOT load upfront; trigger on demand at Gate entry.
CHT Framework
Three dimensions. Score each 1–5 after completing the draft.
C — Completeness
Does the document cover everything it should?
| Score | Meaning |
|---|---|
| 5 | All template sections filled. No [GAP] markers. |
| 4 | All sections filled. 1–2 minor [GAP] markers with justification. |
| 3 | Most sections filled. Some [GAP] markers. Core content present. |
| 2 | Significant sections missing or skeletal. Multiple [GAP] markers. |
| 1 | Document is a stub. Insufficient context was provided. |
Checks:
- Every section in the template/skeleton is addressed
- Required structural markers are present (at least one
[DECISION]in ADRs, at least one[ACTION]in runbooks) - Cross-references between sections are consistent
- If a section is intentionally omitted, the omission is justified
H — Helpfulness
Can the target audience understand and act on this document?
| Score | Meaning |
|---|---|
| 5 | Reader can act immediately. Clear next steps, no ambiguity. |
| 4 | Reader understands fully. Minor clarification might help in edge cases. |
| 3 | Reader understands the main points. Some sections need more context. |
| 2 | Reader would need to ask follow-up questions to act. |
| 1 | Document is confusing or disorganized. Audience mismatch. |
Checks:
- Technical depth matches stated audience
- Jargon is appropriate (defined if cross-functional audience)
[ACTION]items have enough context to be actionable- Document structure follows a logical reading path
- Conclusions and summaries appear where readers expect them
T — Truthfulness
Is every claim grounded in source material?
| Score | Meaning |
|---|---|
| 5 | Every claim sourced. No inference without labeling. |
| 4 | Claims sourced. 1–2 minor inferences clearly labeled. |
| 3 | Most claims sourced. Some inferences not explicitly labeled. |
| 2 | Several unsourced claims. Risk of misleading reader. |
| 1 | Significant fabrication or unsupported assertions. |
Checks:
- Every factual statement traces to a source artifact
- Inferences are labeled ("Inferred:", "Likely:")
[DISPUTED]markers present where sources conflict- No invented rationale attributed to people or teams
- Error messages, commands, file paths are exact (not paraphrased)
- If the source artefact arrived with an
*.envelope.jsonsidecar, the provenance block records the envelope'smessage_id,thread_id,from,performative, andverify_pass/verify_failoutcome. Averify_faildoes not lower the score below 4 by itself; it is captured as[DISPUTED]in the document.
Gate Decision
| Condition | Action |
|---|---|
| All dimensions ≥ 4 | DELIVER — document is ready |
| Any dimension 2–3 | REVISE — one targeted revision pass on failing dimensions only |
| Any dimension 1 | ESCALATE — insufficient context to produce this document. Report what's missing. |
Revision Protocol
When revising:
- Identify the specific sections causing the low score
- Fix only those sections — do not rewrite the entire document
- Re-check the specific dimension that failed
- Deliver with a note on what was revised
Hard cap: one revision pass. If the document still has issues after revision, deliver it with the remaining issues explicitly flagged in the provenance metadata. The requester can provide additional context for a follow-up pass rather than the Scribe entering an unbounded loop.
Provenance Metadata
Every delivered document includes a provenance block at the end:
---
## Provenance
- **Scribe version**: <version>
- **Document type**: [type]
- **Generated**: [timestamp]
- **Source artifacts**:
- [list of source artifacts used, with identifiers]
- **CHT scores**: C:[N]/5 H:[N]/5 T:[N]/5
- **Coverage**: [brief assessment — what's well-covered vs gaps]
- **Flags**: [any unresolved `[GAP]` or `[DISPUTED]` markers]
Source Artifact Identifiers
Use the most specific identifier available:
| Source Type | Identifier Format |
|---|---|
| Git commit | commit:abc1234 |
| File | file:path/to/file.rb |
| Agent output | agent:[agent-name]:[task-id] |
| Conversation | conversation:turn-[N] |
| External doc | doc:[title-or-url] |
| Spec / ticket | spec:[ID] |
| ECL envelope | ecl://thread/<thread_id>/message/<message_id> |
Memory Ingest (Phase G — after DELIVER)
Once the Gate decision is DELIVER (or DELIVER-with-flags after revision), persist the document handoff to CRYSTALIUM.
Ingest
If the document was handed off with an ECL sidecar envelope (*.envelope.json),
ingest via:
mcp__crystalium__ingest(
envelope = <the validated *.envelope.json contents>,
payload = <delivered document contents>
)
This records the document at T1 (from.eidolon=idg drives tier derivation).
If no ECL sidecar is present, commit an episodic note directly:
mcp__crystalium__commit(
layer = "episodic",
payload = <document type + CHT scores + key decisions/gaps summary>,
provenance = { author_agent: "idg", document_type: <type> }
)
author_agent MUST be "idg" on every direct commit.
Session end
After ingest (or after delivery if CRYSTALIUM is absent), call:
mcp__crystalium__session_end()
This triggers Dream consolidation asynchronously. Call it once per IDG dispatch completion. Dream promotes corroborated episodic entries (terminology conventions, structural patterns) to the semantic layer — improving future recall quality.
Graceful skip: if mcp__crystalium__* tools are unavailable (CRYSTALIUM not
installed), skip ingest and session_end and mark the Gate phase complete normally.
Never hard-fail on absent CRYSTALIUM tools. IDG is EIIS-standalone-conformant.
Scribe — Verification Skill
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.