Methodology skill. Given a candidate act of Trump-administration state action, produce a scored, gated, sourced event record for the Retribution Accountability Tracker. This is the reasoning companion to the retribution_events.schema.json data model. Canonical rules live in docs/retribution-methodology.md — this skill operationalizes them.

Step 1 — The retribution gate (threshold)

Ask: was the act motivated by the target’s protected conduct, speech, identity, political association, or legal representation — rather than legitimate, evenhanded governance?

Score the gate by counting indicators (methodology §1):

• Stated retaliatory intent (actor or President tied the action to disfavored conduct)
• Selective application (applied to a critic but not to similarly situated allies)
• Temporal proximity to the target’s protected act
• Absence of a legitimate baseline (see Step 4)
• Judicial finding of retaliatory / unconstitutional purpose

Decision: 0 indicators → do not log. 1 indicator → log at confidence: low. 2+ → proceed toward credibly_reported / documented. Log a one-sentence gate justification.

Step 2 — Classify the instrument (act_type)

Pick the single primary instrument from the taxonomy (methodology §2): firing_removal, prosecution, subpoena, investigation, clearance_revocation, contract_cancellation, funding_freeze, license_bar_targeting, immigration_weaponization, access_revocation, regulatory_pressure, executive_order. A multi-instrument campaign against one target = multiple events, one per discrete act. Record the specific mechanism in instrument.

Step 3 — Retribution Severity Score (RSS 1–5)

Record rss and a one-line rss_rationale. RSS is a transparent editorial judgment, not false precision.

Step 4 — Attribution baseline (counterfactual)

State in attribution_basis the baseline against which the act reads as retributive (methodology §5): evenhanded prosecutorial discretion, ordinary at-will turnover, the prior grant trajectory, threat-based protection decisions. No baseline → no headline retribution claim (may still log at confidence: low).

Step 5 — Harm dimensions

Score each on none/low/moderate/severe, with a one-line note where non-zero: career, economic_security, reputation, family, chilling_effect.

Step 6 — Documented cruelty (moral dimension)

Only if a verbatim, sourced on-the-record quote evidences gratuitous / punitive intent beyond the instrumental act, add a cruelty_evidence[] entry {quote, speaker, date, source}. Empty otherwise — this field is evidence, never editorializing.

Step 7 — TRC classification

• A — plausible criminal exposure (obstruction, civil-rights deprivation under color of law, abuse of process) → prosecution track
• B — abuse of lawful authority warranting reform / disqualification / professional accountability → institutional-reform track
• C — chilling / symbolic, documented for the record → truth-telling track

Label it a provisional editorial judgment for sequencing accountability work — never a legal conclusion of guilt.

Step 8 — Status, sources, confidence

• status: confirmed (act documented) / estimated (details approximate) / contested (administration offers a counter-rationale → must fill counter_rationale).
• sources[]: at least one at news tier or better; prefer primary_official (court dockets, EO text, IG reports).
• confidence: high only when the act is documented.

Output

Emit a schema-valid event object (see retribution_events.schema.json). Hand attribution of responsible[] to the retribution-responsibility-tracer skill — RAF scores the act; the tracer attributes the actors under the naming rule.

Composes with

retribution-responsibility-tracer · malice-evaluator · patriot-retribution-tracker · legal-research-specialist · democratic-health-monitoring