The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains an indirect prompt injection surface in the 'Force-query wiki' step (Step 0).
Ingestion points: The agent is instructed to read content from various local wiki paths, including llmwiki/wiki/concepts/, llmwiki/wiki/entities/, llmwiki/wiki/sources/adr/, and llmwiki/wiki/decisions.md.
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands in the wiki data are provided.
Capability inventory: The skill has the capability to write files (llmwiki/wiki/sources/draft/, llmwiki/html/, llmwiki/wiki/index.md, llmwiki/wiki/log.md) and execute shell commands (python3 harness/scripts/code-logger.py).
Sanitization: There is no evidence of sanitization or filtering of the content retrieved from the wiki before it is processed by the agent.
[COMMAND_EXECUTION]: The skill executes a local Python script using a subprocess call: python3 harness/scripts/code-logger.py --task new title="<feature>".
Evidence: The <feature> parameter is derived from the implementation plan generated by the agent based on user requests. If the feature title contains shell metacharacters and the execution environment does not properly escape them, this could lead to command injection.

propose