audit

Screenplay audit

You are orchestrating the screenplay-prep audit suite for the screenplay project in the current working directory. The agents are project-parameterized: they read ${CLAUDE_PROJECT_DIR}/Claude Docs/PROJECT_PROFILE.md §0 for the per-screenplay configuration and otherwise behave generically.

Preflight

1. Confirm ${CLAUDE_PROJECT_DIR}/Claude Docs/PROJECT_PROFILE.md exists. If it does not, tell the user this project has not been onboarded yet — they need a PROJECT_PROFILE.md (copy ${CLAUDE_PLUGIN_ROOT}/templates/PROJECT_PROFILE.template.md and fill the «per-project» fields). Do not fabricate parameters; stop here.
2. Read §0. Note which capabilities are available for this project: scene_markers.expected, canonical.facts_doc (null → fact-drift has nothing to audit), locations.master_set (null → no location reconcile), cast_registry (empty → cast-cue skips roster reconciliation), and the §6 motif inventory (empty → motif-tracker reports "nothing to verify"). An absent capability is not a failure — it scopes the run.

Scope

$ARGUMENTS selects the run. With no argument, run the default set: slugline-location-linter, cast-cue-linter, continuity-checker, motif-tracker, fact-drift-auditor, screenplay-fidelity-auditor. Recognized argument tokens (space- or comma-separated): the agent names above, plus dialogue (dialogue-voice-auditor), structure (structure-beat-mapper), rights (rights-clearance-scanner), pitch (pitch-consistency-auditor), proof (proofreader), or all (every auditor).

Run

Dispatch each selected agent (they are read-only — safe to run in parallel). Each returns a structured report. Then synthesize one consolidated summary:

• Group findings by severity the agents assign (P1/P2/P3 or CONTRADICTED / DRIFT / UNSUPPORTED, etc.).
• Lead with the single highest-risk finding and the agent that found it.
• For each agent, give a one-line verdict (clean, or N findings) so coverage is visible — never let a skipped/empty-capability agent read as "passed."
• Note any capability that was absent in §0 (and therefore not checked), explicitly.

Do not edit any file. These are flag-only audits; the writer decides and fixes. If asked for fix routes, point at SCRIPT_REVISION_NOTES.md conventions, but the edits are the human's call.