plur_tensions
Idempotent
Manage contradictions between memory engrams. List, scan, confirm, dismiss, or resolve tensions with LLM-powered detection.
Instructions
Tension lifecycle (#181). Default: list persisted tension records (unresolved first). scan:true runs an LLM contradiction scan, persists NEW detections as records, and skips already-recorded pairs. Lifecycle actions: action:"confirm" (real conflict), action:"dismiss" (false positive — pair suppressed from future scans), action:"resolve" + winner: (loser engram retired). Scan requires OPENAI_API_KEY or OPENROUTER_API_KEY env var, or explicit llm_base_url + llm_api_key args.
Input Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| id | No | Tension record id (T-YYYY-MMDD-NNN) for action mode | |
| scan | No | Run an active contradiction scan using an LLM judge. New detections are persisted as tension records; recorded pairs (any status) are skipped. Requires OPENAI_API_KEY or OPENROUTER_API_KEY env var, or explicit llm_base_url + llm_api_key args. | |
| scope | No | Filter by scope | |
| action | No | Lifecycle action on a persisted tension record (requires id). confirm: mark real. dismiss: false positive, suppress the pair. resolve: pick winner (requires winner), the losing engram is retired. | |
| domain | No | Filter by domain prefix | |
| status | No | List-mode status filter. Default: unresolved records (detected + confirmed). | |
| winner | No | Engram id that wins the tension (action:"resolve" only). The other engram is retired. | |
| persist | No | Persist scan detections as tension records (default true). Set false for a dry-run scan that also ignores the recorded-pair suppress list. | |
| llm_model | No | Model name for scan mode (default: gpt-4o-mini) | |
| max_pairs | No | Maximum candidate pairs to evaluate in scan mode (default: 50) | |
| batch_size | No | Pairs judged per LLM call in scan mode (default: 5). Set to 1 for sequential single-pair judging. | |
| llm_api_key | No | API key for the LLM (scan mode) | |
| llm_base_url | No | OpenAI-compatible API base URL for scan mode (e.g. https://api.openai.com/v1) | |
| min_confidence | No | Minimum confidence threshold for scan mode (0–1, default: 0.7) | |
| temporal_discount | No | Multiply judge confidence by a days-apart ladder (same day x1.0 ... 15+ days x0.3) in scan mode (#240). Overrides the config default (tensions.temporal_discount, off by default). The judge prompt already carries recorded dates; enable this only when date-aware judging alone leaves too many temporal-evolution false positives. |