Social Content Safety Auditor

Social Quality Auditor

Run one ECHO profile on one unit. The asset-gate audits a publish package and its governing records; program-maturity profiles audit a channel portfolio/window. Social outcomes remain measured metrics, not part of either rubric score.

When This Must Trigger

• Before queued posts/assets publish or a paused queue resumes.
• When channel truth, claims, disclosure, engagement manipulation, UGC permission, or reported denominators are in doubt.
• When the user requests a community/B2C/founder program-maturity baseline or rerun.

Quick Start

Run ECHO asset-gate on this five-post batch against channel/claims/UGC records.
Audit the last 90 days with program-maturity-community; report outcomes separately.

Skill Contract

Reads: one asset package or one channel portfolio/window, current governance projections, and dated evidence. Writes: only a permissioned v3 artifact. Done when: one profile is complete or its Unknowns are explicit, with no publication, queue, registry, or engagement side effect.

channel-registry, offer-claims-registry, and permission/consent state own truth. This skill judges frozen state and never manufactures permissions or updates canonical records.

Data Sources

Need	Preferred evidence
Channel truth	Current channel projection, account control, dated rules
Asset craft	Exact rendered/caption/package version and narrative canon
Claims/disclosures	Approved claims plus market/platform relationship facts
Hosting/rights	Permission events, advocacy evidence, response/moderation logs
Observability	Native exports, GA4/GSC, denominator definitions, locked panels
Public signal	Compliant public connectors; adjacent sources labeled proxy

Instructions

Runtime Reads

• ../../../references/auditor-runbook.md
• ../../../references/scoring-semantics.md
• ../../../references/echo-benchmark.md
• ../../../references/runtime-invocation.md
• references/auditor-runtime.md

Runtime and Setup

Read ../../../references/auditor-runbook.md, scoring-semantics.md, echo-benchmark.md, and the ECHO catalog entry. Standalone installs use bundled immutable references/auditor-runtime.md; never fetch mutable main. Before deterministic calls, follow runtime-invocation.md, resolve AARON_SKILLS_ROOT="${CLAUDE_PLUGIN_ROOT:-$(git rev-parse --show-toplevel 2>/dev/null || true)}", and require the scorer, validator, and typed catalogs. If unavailable, return score_state: NOT_SCORED / score_confidence: not_scored with no gate verdict or persistent artifact.

Declare target, profile, assessment_mode, program_archetype, channels, market, window, and observation date. Use assessment_mode: asset plus program_archetype: not-applicable for asset-gate; use assessment_mode: program plus the exact community|b2c|founder archetype for a maturity profile. Profile and context must match.

Profile Procedure

• asset-gate: score E1, C1–C10, H1/H2, and O1. An observed asset with no performance rate/metric claim passes O1; missing asset access is Unknown.
• program-maturity-community: score E/H/O with community weights.
• program-maturity-b2c: score E/H/O with feed-led weights.
• program-maturity-founder: score E/H/O with founder-led weights.

Do not add Craft into program maturity or average asset and program results. Every observed state needs source/date/type/confidence. Closed-platform data requires user export; missing private evidence is Unknown.

Verify profile-relevant vetoes: ECHO-E1 governed channel contradiction, ECHO-C1 material claim failure, ECHO-C2 missing applicable disclosure, ECHO-H1 manufactured/baited engagement, ECHO-H2 unpermissioned UGC republication, and ECHO-O1 hidden/switched denominator or proxy-as-measured. Run the typed scorer.

§2 ECHO Worked Examples

• Complete asset gate, raw 88, no veto/fail: DONE/SHIP, final 88.
• Complete asset gate, raw 82, one verified H2 failure: DONE_WITH_CONCERNS/FIX, final 59.
• Complete asset gate, verified C1 and C2 failures: DONE/BLOCK, no final score.
• Program profile without a native export for a required O item: NEEDS_INPUT/UNDECIDED, no score.

§3 ECHO Guardrails

• Missing channel/permission evidence is Unknown, not automatic failure.
• A native attributed share is not automatically UGC republication; paid/off-platform reuse needs its own permission.
• Genuine feedback questions are not engagement bait; coordinated/rewarded mechanics are judged from evidence.
• Platform posting-hour/hashtag folklore is dated Estimated context, never a veto.
• Qualified reach, pipeline, community retention, and other outcomes are reported with denominators/controls outside the rubric.

§5 ECHO Translation

State asset-gate or exact maturity profile in every result. On trace request, qualify all ECHO IDs, especially ECHO-O1 versus ROAS-O1 and ECHO-E1 versus TALE-E1.

Report and Verdict

Begin with the auditor-runbook's exact typed conversation header. Never replace status, verdict, or score_state with prose; list each explicitly missing qualified item as ``ID: `unknown``` before findings.

Show verdict, unit/profile/context/date, score or coverage/interval, confidence, included-item detail, verified critical controls, Unknown inputs, and fixes. For asset batches, identify the exact asset/location. Never publish or unpause a queue based only on the audit; separate explicit approval is required.

Validation Checkpoints

• One unit/profile/window is declared; asset/program constructs are not mixed.
• Current channel/claims/permission state was used, not a stale candidate file.
• Expected items have valid states and provenance.
• O1 is evidence-backed Pass when an observed asset makes no metric claim; missing access stays Unknown.
• Outcomes are not smuggled into a rubric score; no side effect occurred.

Persistence

Persist only after explicit authorization to memory/audits/social/YYYY-MM-DD-<topic>.md. Preserve the scorer's orthogonal status and verdict; validate the complete v3 draft with validate-audit-artifact.py against the intended relative path, persist only through one full-content Write, and revalidate the target per the auditor runbook. Edit/shell/MCP mutations of the reserved sink are unsupported. Do not autonomously write channel state, permissions, claims, queues, or hot cache.

Reference Materials

• ECHO benchmark
• Auditor runbook
• Scoring semantics
• Measurement protocol

Next Best Skill

• Channel truth: channel-registry
• Claims: offer-claims-registry
• UGC/response: engagement-inbox-manager
• Measurement: social-measurement-loop