The Agent Skills Directory

[COMMAND_EXECUTION]: The skill defines several hooks in its YAML frontmatter (e.g., UserPromptSubmit, PreToolUse, PreCompact) that execute shell commands and Python scripts (plan-hook.py, session-catchup.py). These operations are used to inject planning summaries into the prompt and synchronize session state with the plan.html file.\n- [PROMPT_INJECTION]: The skill implements a mechanism to read from a local plan.html file and inject data into the agent's context window. This creates a surface for indirect prompt injection if the file content is modified by an untrusted source.\n
Ingestion points: Data is read from plan.html and injected into the prompt via the plan-hook.py script during various lifecycle hooks.\n
Boundary markers: The skill uses ===BEGIN PLAN DATA=== and ===END PLAN DATA=== to isolate injected content from the agent's primary instructions.\n
Capability inventory: The skill is configured with access to tools including Write, Edit, Bash, and WebFetch as specified in the allowed-tools metadata.\n
Sanitization: No specific sanitization of the artifact content is described; however, the skill includes a SHA-256 attestation mechanism (/plan-attest) to detect and block injection if the file has been tampered with.

plan-it-ar