agentleFS
astrowickedastrowicked/devils-council/skills/scorecard-schema/SKILL.md

scorecard-schema

Skill2 starsChanged 6 months agoMITView on GitHub

Canonical scorecard contract every persona targets and the Phase 3 review engine enforces. Defines finding fields (target, claim, evidence, ask, severity, category), the 4-tier severity enum, the verbatim-evidence grounding rule, and the structural drop rule for banned vague verbs. Consumed by every persona via the subagent `skills:` preload field.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 0 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Do not use this skill for: defining persona voice (that is skills/persona-voice/SKILL.md in Phase 2 Plan 02), classifying signals (that is …

    line 40
  2. Very specific:

    If a persona has no findings, findings: MAY be empty — but the prose summary below MUST explain why (e.g.

    line 77
  3. Very specific:

    Every entry in findings[] MUST carry all seven fields below.

    line 83

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

scorecard-schema

The load-bearing output contract for every persona in devils-council. This skill is not user-invocable; it is preloaded into every persona subagent via the skills: frontmatter field. The Phase 3 review engine validates every scorecard file produced by a persona against the schema defined here.

Decision anchors: D-01 (markdown + YAML frontmatter with findings: list), D-02 (finding fields: id, target, claim, evidence, ask, severity, category), D-03 (4-tier severity, no praise), D-04 (verbatim evidence, ≥8 chars), D-05 (free-text category), D-06 (templates/SCORECARD.md mirrors this schema), D-07 (banned-phrase structural drop, not advisory). Anchors are defined in .planning/phases/02-persona-format-voice-scaffolding/02-CONTEXT.md.

When to Use

This skill is not invoked directly. It is preloaded into every persona subagent so the persona's system prompt carries the contract it must produce. Every scorecard a persona writes at .council/<ts>-<slug>/<persona>.md MUST conform to the schema below.

The Phase 3 review engine validates scorecards against this schema. Non-conforming output is dropped with a structural error logged, never silently included in council output. This is the single mechanism that makes "non-generic critique" (PITFALLS.md §Pitfall 1) and "voice collapse" (PITFALLS.md §Pitfall 3) enforceable rather than aspirational.

Do not use this skill for: defining persona voice (that is skills/persona-voice/SKILL.md in Phase 2 Plan 02), classifying signals (that is Phase 6), or synthesizing findings across personas (that is the Chair in Phase 5).

Scorecard File Shape

A scorecard is a single markdown file. YAML frontmatter at top carries the structured, machine-validated findings. The markdown body below the frontmatter carries a persona-voice prose summary — advisory context, not the load-bearing contract.

Minimal complete example:

--- # YAML doc delimiter
persona: staff-engineer
run_id: 2026-04-22T14-23-01Z-add-caching-layer
findings:
  - id: "sha256:placeholder"        # deterministic hash — Phase 5 / CHAIR-06 defines the recipe
    target: "src/cache.ts:42-48"    # line reference OR quote anchor
    claim: "Cache eviction policy is implicit and unbounded"
    evidence: |
      42:   const cache = new Map();
      43:   function put(k, v) { cache.set(k, v); }   // no max size, no TTL
    ask: "Declare a max-size and eviction policy, or document why unbounded is safe"
    severity: major
    category: complexity
--- # YAML doc delimiter

## Summary

One-paragraph persona-voice prose here. The persona's frame on the
artifact as a whole. Structured findings above are the load-bearing
part; this summary is advisory context, not the contract.

If a persona has no findings, findings: MAY be empty — but the prose summary below MUST explain why (e.g. "out of scope for this persona"). An empty-findings scorecard with no explanation fails validation.

Finding Fields (required)

Every entry in findings[] MUST carry all seven fields below. Extra fields are silently ignored by the validator; missing required fields drop the finding.

FieldTypeRequiredDescription
idstringyesDeterministic hash of persona + target + claim. Phase 5 (CHAIR-06) defines the hashing recipe; Phase 2 only declares the field exists. Phase 3 engine may emit a placeholder (sha256:placeholder) until the hashing is wired.
targetstringyesLine reference (path:line or path:start-end) for code, heading anchor (## Heading) for plans/RFCs, or a short quote-anchor phrase for unstructured prose.
claimstringyesOne-sentence statement of the specific failure mode. Must name the thing that is wrong; no generic "consider edge cases" language.
evidencestringyesVerbatim substring of INPUT.md, minimum 8 characters. See Evidence Grounding Rule below.
askstringyesConcrete remediation the author can act on. Answers "what should I change?" in a single line or short block.
severityenumyesOne of blocker | major | minor | nit. See Severity Tiers below.
categorystringyesFree-text persona-suggested tag (examples: complexity, observability, cost, business, auth-bypass). No fixed enum in v1. Chair synthesizes top-3 categories across all personas in Phase 5.

Severity Tiers (D-03)

Exactly four tiers. The enum is closed; any other value fails validation.

  • blocker — the artifact should not land in this form. The issue is correctness, safety, or contract-breaking. Named personas can and should use this sparingly; a scorecard that flags everything as blocker has no signal.
  • major — substantial concern the author should address before landing or explicitly accept (e.g., missing observability, unbounded growth, silent failure mode).
  • minor — real issue but not landing-blocking; improves quality if addressed.
  • nit — stylistic or small-scope. Collapsed by default in the Phase 7 severity-tiered output rendering.

There is no praise tier. Adversarial discipline first — a persona is not required to say anything positive, and the schema does not give them a slot to do so. This is a deliberate constraint on voice, per PITFALLS.md §Pitfall 3.

Evidence Grounding Rule (D-04)

The contract that makes non-generic critique enforceable at schema level, not at prompt-engineering level.

  • evidence MUST be a verbatim substring of .council/<ts>-<slug>/INPUT.md (the snapshotted artifact for this run). INPUT.md is the single source of truth for what the persona reviewed; evidence that does not appear in INPUT.md cannot be trusted.
  • Minimum length: 8 characters. Prevents meaningless one-word or one-token evidence like evidence: "the" or evidence: "// TODO".
  • Phase 2 states the rule. The Phase 3 review engine enforces it via a literal substring check (grep -F-equivalent) of the evidence string against INPUT.md. Findings that fail evidence-grounding are dropped from the scorecard with a structural error logged, never silently retained.
  • The evidence block MAY include leading whitespace and line-number prefixes to make quoted code readable (see the cache example above). Normalization policy (whitespace collapsing, line-prefix stripping) is a Phase 3 engine-design concern; Phase 2 declares the rule, not the normalization policy.

Banned-Phrase Drop Rule (D-07)

The structural defense against voice collapse and vacuous critique. This rule is schema-level, not advisory. It is independent of any LLM-based tone check.

  • Each persona declares a banned_phrases list in its agent frontmatter (authored in Phase 3+). Every persona bans at minimum: consider, think about, be aware of — the three phrases PITFALLS.md §Pitfall 1 identifies as the strongest generic-critique markers.
  • A finding whose claim or ask field contains a banned phrase — case-insensitive word-boundary match, e.g. \bconsider\b — is dropped from the scorecard. Not flagged advisorily; not rewritten; dropped. The Phase 3 engine logs a structural error recording which phrase matched.
  • Downstream policy (whether the Phase 3 engine regenerates the persona's scorecard, asks for a retry, or ships the remaining findings with a count of drops) is a Phase 3 engine-design decision — not Phase 2's call. This schema only commits that the drop rule exists and is structural.

Category Guidance (D-05)

category is free-text, persona-suggested. There is no fixed enum in v1 by design: letting each persona surface its own framings is what lets the Chair's top-3 clustering in Phase 5 discover emergent categories that a fixed enum would have pre-committed away.

Non-exhaustive seed list personas may draw from:

complexity, observability, cost, business, auth-bypass, race-condition, iam-wildcard, air-gap, egress, blast-radius

Personas MAY invent new categories when the seed list does not fit. A persona that always reaches for the same category across varied artifacts is a voice-collapse signal (see PITFALLS.md §Pitfall 3) — but that is a Phase 4 blinded-reader concern, not a Phase 2 validator concern.

Template Reference (D-06)

The canonical copy-paste template for a scorecard file is at templates/SCORECARD.md. Copy it to .council/<ts>-<slug>/<persona>.md, fill in the frontmatter, write the prose summary.

The template mirrors this schema exactly. If the two disagree, the schema wins and the template must be updated. Schema drift is caught by Phase 3's validator fixtures exercising the template shape.

Relationship to Codex Deep-Scan (shape compatibility)

skills/codex-deep-scan/SKILL.md already declares that its Response Schema findings[] array conforms to this scorecard finding shape. The two schemas MUST stay in sync. Adding a required field here is a breaking change to both.

One-field delta to resolve: codex-deep-scan currently documents severities as blocker | high | medium | low | nit. Phase 2 canonicalizes the severity enum to blocker | major | minor | nit (no high / medium / low). Phase 6, when it wires Codex delegation into personas, owns the mapping from Codex output severities into this canonical enum. One reasonable mapping:

  • high → major
  • medium → minor
  • low → nit

This mapping is documented here as a note; it is not implemented in Phase 2 and skills/codex-deep-scan/SKILL.md is not edited by this plan. Phase 6 is the correct place to wire the translation and to either update codex-deep-scan or implement the mapping at the conductor layer.

Revision Policy

Schema changes require a documented revision, not a silent edit. Adding an optional field is backward-compatible; adding a required field, renaming a field, or removing a field is a breaking change. The scorecard schema and templates/SCORECARD.md MUST stay in sync — drift between them silently weakens validation and must be caught by Plan 03's validator fixtures.


Contract owner: devils-council Phase 2 (02-01) Consumers: all personas authored from Phase 3 onward; Phase 3 review engine validator; Phase 5 Chair synthesis; Phase 6 Codex delegation.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.