Skip to content

donaldfilimon/abi / .agents/skills/abi-superpower-constitution

abi-superpower-constitution

Verified

Constitution governance superpower. 6-principle response audit with E-score, veto, and surfaced telemetry.

ABI Superpower: Constitution

Exposes the 6-principle constitutional audit as a superpower. Observability-only, not a gate — sets audit_passed/audit_vetoed/escore in metadata and logs warnings, but complete/run still return the response.

Real access path

ABI does not expose /abi-superpower-constitution or a standalone constitution CLI command. The audit runs inside completion flows; exercise it with abi complete <input> or the MCP ai_complete / ai_run / ai_learn tools and inspect the returned/stored metadata:

  • audit_passed (bool)
  • audit_vetoed (bool) — hard veto if safety OR privacy < 0.5
  • escore (f32) — weighted constitutional score
  • Per-principle scores (0.0-1.0)

Direct per-principle evaluation is a Rust library/test surface, not a public slash command.

The 6 Principles

PrincipleDescriptionVeto Class
truthfulnessFactual accuracy, no hallucination—
safetyNo harm, violence, illegal actsSafety (hard veto if < 0.5)
helpfulnessUseful, actionable, complete—
fairnessNo bias, discrimination—
privacyNo PII, secrets, confidential dataSafety (hard veto if < 0.5)
transparencyClear about limitations, sources—

Scoring Mechanics

  • Substring matching (infix, case-insensitive) — "harm" fires on "harmless"
  • 7 hardcoded negative substrings per principle — cannot detect novel patterns
  • Weighted E-score — aggregates principle scores with configurable weights
  • Hard veto — if safety < 0.5 OR privacy < 0.5, audit_vetoed = true

Surfaced Telemetry

When store_result=true in completion:

  • audit_passed (bool)
  • audit_vetoed (bool)
  • escore (f32)

MCP tools ai_complete/ai_run/ai_learn include audit fields in response.

Implementation

ComponentSource
Constitution Corecrates/abi-ai/src/constitution.rs — validate(), evaluateResponse(), AuditResult
Completion Integrationcrates/abi-ai/src/completion.rs — audit called post-generation, metadata stored
MCP Toolscrates/abi-mcp/src/ai_tools.rs — audit fields in tool responses

Claim Boundary

Per docs/contracts/external-claims-audit.mdx and docs/spec/multi-persona-technical.mdx:

  • ✅ 6-principle governance validation with per-principle scores
  • ✅ Weighted E-score and safety/privacy lexical veto telemetry
  • ✅ Surfaced in completion metadata and MCP responses
  • ❌ NOT a gate — responses still returned even on veto
  • ❌ NOT novel-harm detection or a general safety classifier; the current implementation is a small lexical rule set
  • ❌ NOT case-sensitive — infix substring match ("harm" matches "harmless")
  • ❌ NOT regulatory certification — repo has no compliance evidence
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