Skip to content

gustavo-meilus/superpipelines / plugins/superpipelines/skills/pipeline-auditor-protocol

pipeline-auditor-protocol

Verified

Loaded by the pipeline-auditor agent to audit Superpipelines pipeline bundles against compliance, CAD, topology, and bundle hygiene rules.

Pipeline Auditor — Operational Protocol

Operating Modes

<operating_modes>

ModeTriggerScope
FULLDirect invocation or first audit.Entire pipeline bundle: all agents, skills, and topology.
DELTATriggered by mutation commands.Changed files plus immediate neighbors and the entry skill.
SCOPE-WIDEaudit-steps --all command.FULL audit across all registered pipelines in all scope roots.
</operating_modes>

Workflow

2. AUDIT

  • Detect layout FIRST: determine whether the bundle is data-only (CAD agents + entry.md under .superpipelines/pipelines/{P}/) or legacy old-root (zero-body agents + run-{P} skill under tool dirs), per references/compliance-matrix.md § "Pipeline layout & criterion applicability". Apply criteria by layout: legacy layout/frontmatter/Lean-Agent criteria (1, 4, 5, 8, 9, 10, 10a, 19) are N/A on data-only — CAD-01..CAD-10 govern there instead. Never FAIL a conformant CAD against a legacy-only criterion.
  • Compliance Matrix: Execute the full compliance check in references/compliance-matrix.md (criteria 1-25 including 10a, resolver consolidation criteria PR-01..PR-05 and PR-07..PR-10, canonical agent-def criteria CAD-01..CAD-10, and bundle-maintenance criteria BUNDLE-01..BUNDLE-07 when auditing the Superpipelines bundle), honoring the layout applicability table.
  • Topology Rules: Verify graph validity, agent coverage, and edge consistency via references/topology-rules.md.
  • Assign severity per references/severity-classification.md and select fixes from references/fix-templates.md.

3. REPORT

  • The auditor is read-only (disallowedTools: Write, Edit, Bash) and NEVER writes the report file or mutates registry.json. Persistence is the orchestrator's responsibility (see commands/audit-steps.md REPORTING).
  • Render the full report per references/audit-report-template.md as inline output in the agent's response, together with the executive summary.
  • Hand the orchestrator an explicit registry-update instruction: the target audit/latest.md path and the last_audit timestamp value to record.
  • Record every audit, even those with zero findings.

4. FIX ROUTING

  • Auditor remains read-only.
  • SEV-0/1: Route to pipeline-architect with the remediation plan.
  • SEV-2/3: Surface to the user for manual decision.

Model-Tier Resolution Criteria (v2.0)

IDCriterionSEVDetection
MT-01Hardcoded model ID in skill bodySEV-2grep -E "claude-(sonnet|opus|haiku)-[0-9]|gpt-5\.[0-9]|gemini-3\." skills/**/SKILL.md skills/**/references/*.md returns matches outside skills/sk-platform-dispatch/profiles/
MT-02Agent missing both model_tier: and model:SEV-1Agent frontmatter has neither field. Runtime resolver tolerates (defaults to fast) but scaffold-time auditor blocks: explicit declaration required for v2.0+ agents.
MT-03Agent has explicit model: without comment justificationSEV-3Escape hatch in use. Surface to reviewer; do not block. Distinguish by plugin_version: if absent or < 2.0.0 → v1 legacy (Phase 0.45 migrates automatically); if >= 2.0.0 → intentional v2 escape hatch (advisory only).
MT-04Profile JSON missing model_tiers_version fieldSEV-2Required for drift detection.
MT-05Preference file references a model not in any profile's catalogSEV-2Compare every prefs.platforms[*].tiers[*] value against the union of all profiles' model_tiers[*].model. Mismatch likely typo or stale ID.
MT-06Agent has effort_tier: set on a platform with effort_field_name == nullSEV-3Effort will be silently ignored on this platform — inform user. Detection requires knowing the source/runtime tier.

Resolution

  • MT-01: Move the hardcoded ID to a profile JSON; reference via platform_profile.model_tiers[tier].
  • MT-02: Add model_tier: to the agent (architect should have done this in Phase 4).
  • MT-03: Add a comment line above model: documenting WHY the escape hatch is needed.
  • MT-04: Add "model_tiers_version": "YYYY-MM-DD" to the profile.
  • MT-05: Run /superpipelines:change-models Mode F (catalog refresh) to reconcile.
  • MT-06: Either drop effort_tier: or accept that it's a no-op on this platform.

Reference Files

  • ${CLAUDE_PLUGIN_ROOT}/skills/pipeline-auditor-references/references/compliance-matrix.md
  • ${CLAUDE_PLUGIN_ROOT}/skills/pipeline-auditor-references/references/canonical-agent-def.md
  • ${CLAUDE_PLUGIN_ROOT}/skills/pipeline-auditor-references/references/topology-rules.md
  • ${CLAUDE_PLUGIN_ROOT}/skills/pipeline-auditor-references/references/severity-classification.md
  • ${CLAUDE_PLUGIN_ROOT}/skills/pipeline-auditor-references/references/audit-report-template.md
  • ${CLAUDE_PLUGIN_ROOT}/skills/pipeline-auditor-references/references/fix-templates.md
rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first.Stars: 194kUpdated Claude CodeCodexVerifiedInstall