The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the ax (axctl) command-line interface to retrieve metadata, register new plans, and update experiment statuses. This is the intended primary function of the skill, leveraging a local CLI tool to manage the Agent Experience (ax) pipeline.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from retrospective logs and investigation prompts provided by the local database. While this data shapes the agent's reasoning, it originates from the user's own environment and requires user approval for any resulting actions.
Ingestion points: JSON output from ax retro meta containing retros and investigation_prompts fields.
Boundary markers: None explicitly present in the data processing step; the agent processes the JSON output as trusted structured input.
Capability inventory: Includes the ability to execute ax CLI commands and read/write files in ~/.claude/ and ~/.agents/ directories.
Sanitization: Not explicitly implemented; the skill assumes the integrity of the local ax database and the human-in-the-loop requirement provides a final check.
[DATA_EXPOSURE]: The skill accesses local configuration files and skill definitions in directories such as ~/.claude/skills/ and ~/.agents/skills/ to perform its analysis. It also reads project-specific guidance files. No evidence of data exfiltration to external domains was detected, and operations are confined to the local filesystem.

retro-meta