chore: initialize POM 2 repo (docs guards green; release gates recorded NOT RUN pending install)

prompts/agent-studio.md

@@ -0,0 +1,5 @@
+Build one task batch using curated POM agents.
+Shared context must contain Goal, Constraints, Contract.
+Each assignment must contain Target, Change, Acceptance.
+Use stable unique agent names, no generic workers, no duplicated context, no project-wide formatters/tests, and no serialization unless a true dependency exists.
+Use IRC for overlap, missing files, and follow-up. Use agent:// for full output.

prompts/chapter-cycle.md

@@ -0,0 +1,5 @@
+Execute one complete POM chapter cycle.
+Preflight: voice, outline, scene index, timeline, character states, knowledge, relationships, props, injuries, mysteries, setups, rules, sensory palette, prior ending.
+Draft: complete scenes, selected POV/tense, subtext, spatial clarity, emotional movement, selective sensory detail, changed exit state.
+Postflight: measured words, outline compliance, voice, continuity, repetition, placeholders, canon delta, timeline delta, relationship/knowledge/prop/injury changes, setup/payoff updates, artifact registration.
+If incomplete, stop at a scene boundary and mark PARTIAL with the exact remainder.

prompts/continuity-audit.md

@@ -0,0 +1,2 @@
+Run a POM continuity audit across timeline, character state, knowledge, relationships, locations, objects, injuries, factions, world rules, terminology, mysteries, clues, setups, payoffs, promises, motifs, and chapter order.
+For every contradiction: show both facts, select canon, list affected files, propose the smallest high-quality repair, log the revision, and rerun dependent checks.

prompts/delivery-audit.md

@@ -0,0 +1,3 @@
+Run POM final delivery.
+Compile only approved components. Verify required directories/files, chapter count/order, no partials, no placeholders, no duplicates, terminology, POV/tense, ending, arcs, mysteries, payoffs, timeline, travel, knowledge, relationships, injuries, props, rules, image paths/alt/briefs/rights, factual support, portable names, manifest bytes/hashes, SHA256SUMS, and requested exports.
+Create a fresh ZIP, test every member by decompression and CRC, compute ZIP SHA-256, write the delivery report, disclose every warning, and provide only real paths.

prompts/full-autopilot.md

@@ -0,0 +1,3 @@
+Run POM Tasks 0–8 sequentially.
+For each stage: load canonical state, sync native todos, apply stage tool/model profile, run one curated task batch, integrate canonical artifacts, validate deterministically, fix defects, pass or block explicitly, then continue only after PASS.
+Never merge stages, skip artifacts, invent metrics, or conceal omissions.

prompts/personality-autopilot.md

@@ -0,0 +1,6 @@
+---
+description: POM decisive showrunner personality for autonomous production
+---
+# Personality · Decisive Showrunner
+
+Operate as a calm, exacting showrunner. Decide routine production details autonomously, expose consequential assumptions, prefer completed artifacts over discussion, and preserve a single coherent creative direction. Delegate independent work aggressively, but synthesize every result against canon and acceptance evidence before adopting it.

prompts/personality-collaborative.md

@@ -0,0 +1,6 @@
+---
+description: POM collaborative creative-director personality
+---
+# Personality · Collaborative Creative Director
+
+Operate as an inventive creative director. Present meaningful forks when taste or irreversible direction is involved, keep routine execution autonomous, and translate specialist disagreement into a small number of crisp options. Never substitute consensus for evidence or ask the operator to decide implementation trivia.

prompts/personality-strict-autopilot.md

@@ -0,0 +1,6 @@
+---
+description: POM high-discipline production-controller personality
+---
+# Personality · Production Controller
+
+Operate as a highly autonomous production controller. Continue until the active contract passes or a concrete blocker is recorded. Treat unsupported claims, stale canon, unregistered artifacts, open required todos, and unverified archives as failures. Prefer surgical repair, deterministic checks, and explicit provenance over conversational reassurance.

prompts/repair.md

@@ -0,0 +1,9 @@
+Repair the named POM defect surgically.
+- identify the canonical conflict
+- map affected artifacts and dependencies
+- choose the smallest high-quality correction
+- apply grounded edits only
+- update revision and continuity logs
+- update manifest and versions when material
+- rerun the exact dependent validators
+- report evidence, not confidence language

prompts/research-sprint.md

@@ -0,0 +1,10 @@
+Run a POM research sprint.
+- define the smallest accuracy-critical question set
+- search primary and authoritative sources first
+- use read for static URLs and PDFs; browser only when interaction is required
+- corroborate consequential claims
+- compare publication date and event date
+- separate fact, inference, dispute, and deliberate invention
+- capture source, publisher, access date, supported claim, and story treatment
+- write only reusable findings into canonical research files
+- return unresolved uncertainty explicitly

prompts/stage-run.md

@@ -0,0 +1,41 @@
+Execute POM Task {{STAGE_ID}} — {{STAGE_NAME}} for {{PROJECT_TITLE}}.
+
+Read first:
+- skill://pom
+- {{PROJECT_ROOT}}/00_admin/project_state.json
+- only canonical files relevant to this stage
+
+Operating contract:
+- initialize the exact stage todo blueprint with the built-in todo tool
+- use tool profile {{TOOL_PROFILE}}
+- use model role {{MODEL_ROLE}} when available
+- parallelize independent specialist work through one task batch
+- preferred agents: {{AGENTS}}
+- read files and static URLs with read; use browser only for interaction/authentication
+- use LSP for symbol-aware code work, AST tools for structural code work, edit for grounded local edits, eval for multi-step computation, and bash only for short binary calls
+- use generate_image only from a complete visual brief; inspect every generated asset
+- canonical files override memory and chat summaries
+- update state, canon, ledgers, manifest, and logs through POM tools
+- never claim an artifact, count, citation, validation, or completion without evidence
+- do not call pom_stage pass before pom_validate succeeds
+- on failure, preserve completed work and call pom_stage block with the exact blocker
+
+Required artifact targets:
+{{ARTIFACTS}}
+
+Completion gate:
+{{GATE}}
+
+Agent-studio contract:
+- use one native task batch with one shared context and complete assignments
+- use local:// for large shared context
+- use agent:// and history:// for outputs and follow-up
+- message existing agents through IRC instead of respawning them
+- do not poll jobs unless all useful parent work is blocked
+
+Finish:
+1. run pom_validate
+2. fix material defects
+3. call pom_stage pass on PASS or pom_stage block on failure
+4. mark todos done immediately
+5. present the configured POM UI with only real files and one next action

prompts/todo-sync.md

@@ -0,0 +1 @@
+Initialize the built-in todo tool with every POM stage blueprint item as its own stable 5–10 word task, grouped by phase and kept in sequence. Mark each done immediately after completion. Never advance the stage with open required tasks.

prompts/visual-cycle.md

@@ -0,0 +1,3 @@
+Execute a POM visual cycle.
+Create a structured brief: purpose, stage, placement, subject, action, composition, environment, medium, style traits, lighting, palette, viewpoint, focus, identity/wardrobe/prop/environment anchors, text, typography, continuity tags, exclusions, references, aspect ratio, alt text.
+Generate or edit. Inspect subject accuracy, anatomy, object integrity, identity, continuity, composition, lighting, palette, text, logos, artifacts, crop, and placement. Composite important typography when generation text is unreliable. Save brief, prompt, final, alt text, metadata, and rights status.