AGENT SKILL

ramza-critic

The maker≠checker critique protocol for a plan at Test (T). Mandatory before Assemble at tier=full (self-approval is mechanically DENIED by ramza-gate), recommended at tier=lite. Mechanical debiasing: strip author identity, anchor findings on ramza-lint/ramza-ears-lint output plus the refine rubric, record identities via ramza-gate critic. Read-only; critiques, never edits.

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
ramza-critic
metadata
{ "methodology": "RAMZA" }
description
The maker≠checker critique protocol for a plan at Test (T). Mandatory before Assemble at tier=full (self-approval is mechanically DENIED by ramza-gate), recommended at tier=lite. Mechanical debiasing: strip author identity, anchor findings on ramza-lint/ramza-ears-lint output plus the refine rubric, record identities via ramza-gate critic. Read-only; critiques, never edits.

03 SKILL.md 正文

RAMZA — Critic Skill (maker≠checker)

Use this skill whenever a plan needs independent critique — mandatory at tier=full before Assemble (ramza-gate advance --to A DENIES entry without a recorded critic), recommended at tier=lite. The critic is never the plan's author: ramza-gate critic mechanically enforces --author != --checker and exits 1 on a self-approval attempt.

Inputs

The critique session works from exactly three artifacts — nothing else self-reported by the author:

  1. The plan (.spectra/plans/<slug>.md) — the artifact under review.
  2. The acceptance criteria (the plan's ## Acceptance Criteria section, or a sibling .spectra/plans/<slug>.acceptance.md) — the EARS-form criteria block.
  3. The state file (.spectra/plans/<slug>.state.json) — tier, phase, the prior gate history (gates[]), and any recorded skips.

The critic does not receive the author's private chat reasoning or a prose "why I chose X" narrative that isn't already in the plan body — only the artifact, the criteria, and the mechanical history in state. That's what "strip author identity" buys: a critique anchored on what the plan actually says, not on the author's framing of it.

The debias procedure (mechanical, in order)

1. Strip author identity

Before evaluating content, remove/ignore anything that identifies or argues for the author's authorship rather than describing the plan itself:

  • Do not read critic.author/critic.checker in the state file before forming an independent judgment — read it only afterward, to confirm the --checker value being recorded is really you.
  • Ignore any frontmatter author:/by: field, git-blame authorship, or inline "I chose X because…" framing that argues for the plan rather than stating what it does. Critique the artifact's claims, not the author's narration of their own reasoning.
  • If the plan's author and the critic are the same identity, stop here — ramza-gate critic will reject the recording in step 3; do not proceed to a self-critique in its place.

2. Evaluate criteria-anchored, not vibes-anchored

Run the mechanical checks FIRST, and anchor every subsequent finding on their actual output — never on a re-derived impression of "does this look complete":

ramza-lint --plan <plan.md> --state <state>
ramza-ears-lint <criteria.md>

Every missing section or EARS violation these print is a finding, verbatim — not a starting point for independently re-litigating structure. Structure is mechanical (that's the whole point of ramza-lint / ramza-ears-lint being grep-level tools); the critic's judgment is reserved for what the tools cannot see — does the approach actually address the risk, is a hypothesis's rejection rationale sound, is a timebox realistic, does a story's executor hint match its actual ambiguity.

Score that content-quality judgment through the refine rubric — never estimated in prose:

echo '{"clarity":<1-5>,"completeness":<1-5>,"actionability":<1-5>,"efficiency":<1-5>,"testability":<1-5>}' \
  | ramza-score --rubric refine --state <state> --cycle <current-refine-cycle>

ramza-score computes the verdict (cycle 1: pass requires every dimension ≥3; cycle ≥2: every dimension ≥4) and appends the result to state.gates[] and the calibration log (ramza-calibration.jsonl) — the critic reports the tool's verdict, never a hand-computed one.

3. Record identities

Once the critique is complete, record it — this is what makes the plan eligible to advance to Assemble at tier=full:

ramza-gate critic --state <state> --author <author-id> --checker <checker-id>

--author and --checker MUST be distinct identities — ramza-gate DENYs author == checker with exit 1. This is the mechanical enforcement of maker≠checker, not a convention to remember. Use stable identifiers (e.g. a model class + session tag, or a human reviewer's name); whatever value goes here is the audit trail's permanent record of who checked whom.

Output shape

The critique report has exactly three parts, in this order:

  1. Verdict — the ramza-score --rubric refine result verbatim (total, per-dimension dims, and pass/fail), plus the ramza-lint and ramza-ears-lint exit status (clean, or the violation count each printed).
  2. Per-dimension findings — one line per rubric dimension (clarity, completeness, actionability, efficiency, testability), each citing the specific plan section or acceptance-criteria ID it concerns. A dimension scored below the cycle's bar MUST have a finding; a passing dimension MAY have one too (e.g. a near-miss worth flagging before the next cycle).
  3. Prescription list — an ordered, actionable list the author applies in Refine (R). Each item names the section/ID to change and the specific fix — never "improve clarity," always "AC-004's THEN is compound; split into AC-004a/AC-004b." This list is what Refine consumes; a critique without a prescription list cannot be acted on.
## Critique — <plan slug>

**Verdict:** ramza-lint <clean | N violations> · ramza-ears-lint <clean | N violations> ·
refine rubric: <pass | fail> (total <X>, cycle <N>)

**Findings**
- clarity (<score>/5): <finding, cites section/ID>
- completeness (<score>/5): <finding>
- actionability (<score>/5): <finding>
- efficiency (<score>/5): <finding>
- testability (<score>/5): <finding>

**Prescriptions**
1. <section/ID> — <specific fix>
2. …

Hard constraints (P0)

  1. Maker≠checker is mechanical, not a convention — ramza-gate critic DENYs --author == --checker. Never role-play both sides in one session.
  2. Structural findings come from ramza-lint / ramza-ears-lint output, never re-derived by re-reading the plan for "missing sections."
  3. Content-quality findings are scored via ramza-score --rubric refine, never estimated in prose.
  4. Every finding below the cycle's bar carries a prescription; a verdict with no prescription list is not a usable critique.
  5. Full tier: no path to Assemble without a recorded critic (ramza-gate advance --to A enforces this directly, see docs/methodology/tiers.md).
  6. READ-ONLY — the critic reviews and records; it never edits the plan itself. Fixes land only through the author's next Refine pass.

See SPEC.md "## T — Test" for how the critic gate fits the cycle, docs/methodology/tiers.md for how it fits the tier table, and docs/methodology/scoring.md "refine (critique rubric)" for the rubric's weights and thresholds.


RAMZA — Critic Skill (maker≠checker)

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