---

name: circuit_recovery_harness description: Build and validate bounded mechanism-faithful recovery artifacts for Transformer Circuits proxy experiments.

Circuit Recovery Harness

Use this skill to assemble a recovery experiment for a conceptual mechanistic interpretability paper when no original trained model is available. It should orchestrate generated circuit-analysis skills, write auditable command and invocation logs, and declare proxy status honestly. Do not use it to bypass module skills with duplicated one-off logic; the harness must call or cross-check the generated scripts.

Inputs

• Attempt directory and module plan.
• Generated skill root containing the circuit skills.
• Runtime handoff that records full-runtime blockers and permitted proxy recovery.
• A deterministic synthetic or real attention-only circuit item.

Outputs

• Experiment plan and source manifest.
• Generated data item and optional training/mechanism trace.
• Recovery result with numeric metric and mechanism checks.
• Experiment command log and generated skill invocation log.

Workflow

1. Read the module-plan recovery target and runtime handoff.
2. If trained paper models are unavailable, declare a soft-mode proxy.
3. Construct a repeated-token circuit item.
4. Invoke logit-lens, QK/OV expansion, path-expansion, and induction-detector scripts.
5. Aggregate metrics and mechanism booleans.
6. Run the recovery experiment validator.

Validation

The smoke test checks that required result fields exist and that proxy mechanism checks are not empty.

Limitations

This skill creates recovery evidence, not a full-scale paper reproduction. Full trained-model recovery must replace the synthetic item when the relevant model assets become available.