The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes the route-events.jsonl file which contains request_snippet fields. These snippets store the first 200 characters of user-provided requests. An attacker could include malicious instructions in a previous request, which are then read and potentially followed by the agent during a trace analysis session. * Ingestion points: SKILL.md (Phase 2) and references/trace-schema.md define reading from ~/.claude/learning/route-events.jsonl. * Boundary markers: The skill uses basic quoting for request snippets in the timeline display ("{request_snippet}") but does not implement comprehensive instruction-bypass protection. * Capability inventory: The agent has access to Bash, Read, Glob, and Grep tools, which could be leveraged if an injection is successful. * Sanitization: Snippets are truncated to 200 characters, which limits the scope but not the presence of malicious instructions.
[DYNAMIC_EXECUTION]: The reference files (references/error-handling.md and references/preferred-patterns.md) include Python scripts executed via shell heredocs to perform diagnostic counts and validation on the log file. These scripts are statically defined and used for internal state checking.
[COMMAND_EXECUTION]: The skill uses the Bash tool to perform file system operations such as checking for the existence of the log file (ls -la) and counting lines (wc -l).

explanation-traces