agentsclimarketplace

Credit qa verifier

Skill tywinlu1988/Credence-Global/dev/.claude/skills/credit-qa-verifier

A methodology-first fixed-income credit analysis engine for global markets — delivered as Agent Skills (SKILL.md). Six international paradigms, four-layer architecture, S&P/Moody's/Fitch aligned. Installable in Claude Code, Codex, Cursor, Gemini, and OpenCode.

Install
npx -y skills add tywinlu1988/Credence-Global --skill credit-qa-verifier

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

3 things to look at

  • 21 days oldThe repository was created 21 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use when verifying a fixed-income credit report or analysis — checking a work path's quality gates, enforcing the mandatory signal-density rules (no numeric score below the density floor), the one-shot-veto CCC ceiling, Mode B anti-hallucination guardrails, and single-source-of-truth compliance (no invented thresholds). Triggers on 'QA check', 'review this report', 'check for issues', 'quality gate', or as the final step after report generation. Reads engine documents as the rule source; never relaxes a gate.

SKILL.md

7.9 KB, ~1.7k tokens by cl100k_base, as published. Nobody here has run it

Purpose

Engine version: v0.0.7

Non-Negotiables (see AGENTS.md): no delivery of analysis conclusions without a passing QA Verdict from this skill (pure knowledge questions exempt) · no numbers without a doc §section citation (engine_undefined otherwise) · no report outside dev/templates/.

QA layer — final stage of the four-stage chain. Responsibility: perform pre-delivery review of the Delivery Note and its upstream Analysis Artifact and Path Sheet, producing a QA Verdict. This skill uses engine documents as the rule source and never relaxes a gate: if any quality gate or mandatory check fails, the verdict is fail and must be sent back for remediation; do not lower standards for delivery. This skill does not replicate any thresholds/rating mappings; rule content uses the referenced engine documents as the single source of truth.

Inputs

  • Delivery Note: produced by credit-report-builder (field shape at dev/engine/pipeline-contract.md §2.3).
  • Analysis Artifact: produced by fixed-income-credit-analysis, for reviewing density/veto/completeness (§2.2).
  • Path Sheet quality_gates: produced by credit-analysis-router, serving as the checklist source for gate-by-gate review; path_id is the join key throughout — all three artifacts must be consistent.

Output

  • QA Verdict — final stage artifact, field shape at dev/engine/pipeline-contract.md §2.4. verdict takes one of pass | pass-with-findings | fail.

Verification Protocol

  1. Join key consistency: The path_id in all three artifacts must be identical and resolvable in the registry. Inconsistency → fail.
  2. Gate-by-gate review: Review each quality gate in the Path Sheet's quality_gates list, producing gate_results (each with status + evidence, evidence citing engine document sections).
  3. Four mandatory checks: See below. Any failure → fail.
  4. Produce verdict: All pass → pass; pass but with findings that should be noted → pass-with-findings; any failure → fail with remediation.

Mandatory Checks (rule source is engine documents)

  • Signal density rule density_rule: Dimensions below the density floor must not output numeric scores and must be annotated as insufficient information to evaluate; when weighted-average density is insufficient, must not output a final letter rating. Rule source: dev/engine/mosaic-engine.md §4.3.
  • One-vote veto ceiling veto_ceiling: Issuers triggering a one-vote veto have their rating ceiling locked at CCC, may not be raised. Rule source: dev/engine/industry-framework.md §5.
  • Mode B anti-hallucination mode_b: Unless the user explicitly provides data sources (CSV/API/MCP), no Mode B external data values may appear; all Mode B fields must be treated as data gaps. Rule source: dev/engine/mosaic-engine.md §6.
  • Single source of truth single_source: Reports/analyses must not fabricate thresholds, weights, or rating mappings; quantities not defined in the engine must be annotated as engine_undefined. Rule source: all referenced engine documents.

Process-Compliance Checks (anti-drift; rule sources as noted)

  • Template compliance: Every rendered report file maps to a template in the path's registry templates field (or declared marker) — no ad-hoc layouts. Rule source: dev/engine/work-path-registry.md + dev/templates/index.yaml.
  • Citation compliance: Every numeric claim (threshold, weight, score, tier, rating) carries a doc §section citation or is marked engine_undefined. Rule source: AGENTS.md Non-Negotiables.
  • Dimension compliance: All analysis dimensions/metrics use engine vocabulary only (industry-framework D1-D10 + paradigm pyramids; concentration-framework five dimensions; contagion-matrix 19 industries; P1-P6 paradigms) — no invented dimensions or industries. Rule source: dev/engine/industry-framework.md + dev/engine/contagion-matrix.md.
  • Chain compliance: A Path Sheet exists for the path_id, and this QA Verdict is produced before delivery — no analysis ships without it. Rule source: dev/engine/pipeline-contract.md.

QA Verdict Output

Template (schema single source of truth is dev/engine/pipeline-contract.md §2.4):

path_id: ""                 # join key (all three artifacts must be consistent)
verdict: ""                 # pass|pass-with-findings|fail
gate_results:               # gate-by-gate review results
  - gate: ""                # "rule name (dev/engine/<doc>.md §section)" (inherited from Path Sheet quality_gates)
    status: ""              # pass|fail
    evidence: ""            # review evidence (citing engine document sections, not copying values)
mandatory_checks:           # four mandatory checks
  density_rule: ""
  veto_ceiling: ""
  mode_b: ""
  single_source: ""
remediation: []             # remediation suggestions for failed items

Example (Credit Selector single target WP-CS-01, three gates all pass, mandatory checks all pass, one finding to note):

path_id: WP-CS-01
verdict: pass-with-findings
gate_results:
  - gate: "Signal Density (dev/engine/mosaic-engine.md §4.3)"
    status: pass
    evidence: Dimensions below density floor are set to null and annotated insufficient information to evaluate (mosaic-engine §4.3)
  - gate: "Veto (dev/engine/industry-framework.md §5)"
    status: pass
    evidence: No one-vote veto triggered; rating ceiling rule verified (industry-framework §5)
  - gate: "Cross-Validation (dev/engine/dual-track-methodology.md §4)"
    status: pass
    evidence: Track divergence presented and interpreted in report (dual-track-methodology §4)
mandatory_checks:
  density_rule: pass
  veto_ceiling: pass
  mode_b: pass
  single_source: pass
remediation: []

Chaining (Final Stage)

  • Upstream (REQUIRED): credit-report-builder — consumes its Delivery Note and upstream artifacts.
  • Final stage: This skill is the last of the four-stage chain; no downstream. When verdict is fail, return to the appropriate stage per remediation (density/veto issues → return to analysis; template/assembly issues → return to report) for remediation and re-review.

Guardrails

  • Never relaxes a gate: Quality gates and mandatory checks only have pass/fail outcomes; no discretionary pass. Missing completeness report, outputting numeric scores below density floor, fabricating thresholds, Mode B hallucination — all result in fail.
  • Do not replicate engine content: Only reference rule names and document sections; do not replicate any thresholds, SRI tiers, layered time budgets, or rating mappings. Numerical judgments are based on the referenced engine documents.
  • Engine documents as rule source: Every gate rule name must be grep-able in the referenced engine document (traceability at references/qa-checklist.md); do not fabricate rules.

References

  • references/qa-checklist.md — QA checklist (gate rule name → engine document traceability, pointer only)
  • dev/engine/pipeline-contract.md — Four-stage chain I/O contract (artifact schema single source of truth)
  • dev/engine/mosaic-engine.md — Signal density / completeness / Mode B guardrail (density_rule, mode_b rule source)
  • dev/engine/industry-framework.md — One-shot veto rating ceiling (veto_ceiling rule source)
  • dev/engine/work-path-registry.md — Work path registry (quality gate list traceability)

What ships with it: 1 file

5.4 KB alongside SKILL.md

references/

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.