The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes historical data from past turns which may contain untrusted content.
Ingestion points: Systematic reading of the agent-transcripts/ directory within the active workspace.
Boundary markers: The instructions mandate cross-checking patterns across multiple slices of history and require multiple instances of a signal before codifying it into a skill.
Capability inventory: The skill uses the create-skill tool to write permanent instruction files (SKILL.md) to the user's project or home directory.
Sanitization: Employs the ps-unslop skill for prose discipline and requires a final human vibe-check and manual PR review before the skill is landed.
[COMMAND_EXECUTION]: The skill uses local shell commands to manage the lifecycle and version control of the generated skills.
Evidence: Executes git log to find the last edit time of existing skills and uses git worktree and git commit to manage the drafting and submission process.

ps-automate-me