concept
active
concept:harness-activation-failure

Harness Activation Failure

A failure mode where weak-tier models fail to invoke relevant harness artifacts (e.g., skills) during task solving

Neighborhood — ranked by edge-count

Concepts (3)

concept
  • The capability of a task-solving agent to benefit from updated harnesses during task solving
  • Skill-Load Rate
    associated_with
    The fraction of trajectories in which an agent actively loads at least one skill into its context
  • SkillsBench enforcement mechanism that accepts only single-key JSON actions; composite multi-key actions are rejected, preventing skill loading

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

  • A failure mode where even when harness artifacts are loaded, weak-tier models fail to follow their guidance faithfully
  • The capability of an evolver model to produce useful persistent harness updates from execution evidence
  • Activationsconcept0.761
    Internal representations of the model on which probes operate; the method uses activations to rank datapoints.
  • Standard method in mechanistic interpretability that intervenes on activations; VPD flips this paradigm by patching parameters.
  • Agent Harnessconcept0.753
    The external non-parametric context and infrastructure (prompts, skills, memories, tools) through which an LLM is deployed for task execution
  • Natural-language harness artifacts that encode standing behavioral rules, task policies, and reasoning procedures
  • Clamping activations along the Assistant Axis to remain above a minimum threshold (25th percentile), introduced as a stabilization method
  • The conventional approach (e.g., SAEs, transcoders) of decomposing activations into interpretable features.