The Agent Skills Directory

[SAFE]: The skill consists of markdown instructions designed to guide an agent through a post-mortem analysis of its own or another agent's performance. It does not involve any dangerous operations, command execution, or remote resource fetching.
[INDIRECT_PROMPT_INJECTION]: The skill instructions involve processing external data sources (session transcripts and tool logs), which represents a potential ingestion surface for indirect prompt injection.
Ingestion points: The skill explicitly instructs the agent to use the session transcript, tool-call log, or eval trace from a completed run (Section 1).
Boundary markers: There are no explicit instructions to use delimiters or 'ignore embedded instructions' warnings when reading the transcripts.
Capability inventory: The skill mentions using a harness-experiment tool (Section 5) to rerun tasks and verify fixes.
Sanitization: No specific sanitization or filtering of the input logs is described.
[NO_CODE]: The skill contains no code files, scripts, or binaries. All logic is contained within the markdown instructions for the AI agent.

trajectory-review