| name | editing-story-continuity |
| description | Evaluates story outputs with a rubric, blocks weak handoffs, and routes fixes to the responsible agent, and is used when the Continuity & Meaning Editor runs the final quality gate. |
Acts as the story-quality gate that keeps weak, incoherent, or incomplete outputs from reaching the user.
Use this skill when
- Any agent has produced output that must be evaluated before handoff.
- The pipeline needs a pass/fail decision with actionable feedback.
- Story quality, continuity, or emotional coherence is at risk.
- A failing output must be routed back to the correct agent.
Capabilities
- Validates schema completeness and typed JSON structure.
- Scores outputs against a rubric.
- Detects:
- continuity breaks
- emotional mismatch
- trivial choices
- repetitive prose
- unsupported ending logic
- missing fallback behavior
- Routes the fix to the correct agent instead of rewriting the whole scene by default.
- Returns concise, actionable revision notes.
- Protects the MVP from low-quality story drift.
Workflow
- Validate structure first. If required fields are missing, fail immediately.
- Check narrative integrity.
- Does this match the story bible?
- Does it honor the current anchor?
- Does it reflect hidden state and prior choices?
- Check emotional alignment. Does it still feel connected to the user's original tension?
- Check medium readiness. Can the scene survive partial media failure?
- Score the output.
- Decide pass/fail.
- pass → hand off
- fail → route back to the offending agent with precise notes
- Limit retry loops. If the same step fails twice, escalate to the orchestrator.
Suggested rubric
Score each 1–5:
- emotional resonance
- structural integrity
- choice quality
- continuity and consistency
- multimodal readiness
- clarity and polish
Pass rule
- No missing required fields
- No critical failure in continuity or safety
- Average score of 4+ recommended for story-facing output
Example feedback payload
{
"status": "fail",
"offending_agent": "choice_designer",
"scores": {
"emotional_resonance": 4,
"structural_integrity": 4,
"choice_quality": 2,
"continuity": 5,
"multimodal_readiness": 4,
"clarity_and_polish": 4
},
"feedback": [
"Both choices are functionally 'tell the truth now' with different wording.",
"Revise the choice set so each option protects a different value and carries a visible cost."
]
}
Guardrails
- Do not silently fix major creative problems; send them back.
- Do not block a scene for tiny stylistic preferences.
- Story quality is the priority, but momentum matters in a hackathon — be strict on substance, light on cosmetic perfection.
Project references
- AGENTS.md — typed JSON evaluation rule, story-quality priority, clear agent separation.
- docs/PRD.md — agent output evaluation principle, narrative requirements, fallback behavior.
- docs/roadmap.md — MVP needs one polished demo path; use editing to protect that path.
