The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted transcript data from previous conversations to generate new instructions for the agent.
Ingestion points: The skill reads conversation history from host resources, specific transcript directories, and the environment variable $PSTACK_TRANSCRIPTS_DIR.
Boundary markers: The instructions lack requirements for using explicit delimiters (like XML tags) or safety markers when subagents process transcript segments.
Capability inventory: The skill has the ability to write files (SKILL.md), orchestrate subagents, and execute Git commands (worktree, commit, PR).
Sanitization: There is no mention of filtering or sanitizing the content extracted from transcripts before it is used to draft the new skill's prose.
[COMMAND_EXECUTION]: The skill instructs the agent to perform various shell commands to manage the skill lifecycle.
Evidence: The instructions include using git log to check for updates and performing Git operations like worktree, commit, and PR to save changes.
Context: These operations are part of the intended development workflow but involve direct interaction with the host's version control system.

automate-me