The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes session transcripts and historical work to generate retrospective briefs and improvement proposals. If session logs contain malicious instructions, they could potentially influence the agent's summaries or proposed actions.
Ingestion points: The ax retro brief command reads session transcripts to populate retrospective tasks; ax improve list reads proposals derived from previous agent activity.
Boundary markers: The instructions do not specify the use of clear delimiters or instructions to ignore embedded commands when processing transcript content.
Capability inventory: The skill can execute ax CLI commands, write files (specifically scaffolding SKILL.md files), and dispatch subagents for retrospective review.
Sanitization: There is no evidence of content sanitization for processed transcripts prior to analysis by the agent or subagents.
[COMMAND_EXECUTION]: The skill performs multiple shell command executions via the local ax CLI to manage the experiment graph and project state.
It uses commands like ax improve list, ax improve accept, and ax improve verdict to query and update local project metadata.
These operations are essential for the skill's purpose as a management interface for the ax toolchain.
The instructions explicitly assume the ax binary is already present in the user's environment.

retro