The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by processing historical conversation data to generate new instructions.
Ingestion points: Conversation history is retrieved using t3_thread_read (Step 1) to identify behavioral patterns.
Boundary markers: The skill does not explicitly define delimiters or warnings to ignore embedded instructions within the mined content before it is clustered and codified.
Capability inventory: The skill possesses the ability to write to the local filesystem and manage directory structures across multiple provider paths (Step 4).
Sanitization: While the 'unslop' skill is applied for prose discipline (Step 5), there is no specific sanitization mentioned to filter out adversarial instructions from the mined data.
[COMMAND_EXECUTION]: The skill uses shell commands to automate the installation of generated skills into various environment-specific directories.
Evidence: Step 4 details a bash loop that executes mkdir -p and ln -sfn to create symlinks for the user's home directory paths, specifically targeted at Claude Code, Codex, Grok, and Cursor skill folders.
[DYNAMIC_EXECUTION]: The skill generates new executable agent instructions (SKILL.md files) at runtime based on the synthesis of mined data.
Evidence: The skill orchestrates pstack-author-skill to draft new skill files that are subsequently deployed into active skill-loading paths (Step 4 and 6).

automate-me