The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a mechanism for the agent to process free-form user input during high-friction interactions to update its internal task plan. While this involves processing untrusted data, the skill is specifically designed to mitigate misalignment through a structured feedback loop.
Ingestion points: The skill monitors all user messages during execution phases for frustration signals or corrections (SKILL.md).
Boundary markers: The protocol requires the agent to produce a "numbered, precise, unambiguous summary" (Step 3) which serves as a boundary between the messy user input and the agent's planned actions.
Capability inventory: The skill forces a "Full Stop" of all file edits, code changes, and task executions upon triggering; it does not grant the agent any additional privileges or tools.
Sanitization: The skill relies on an iterative reflection loop (Step 4) and explicit user confirmation (Step 5) to validate that the agent's interpretation of the input is correct before any action is taken.
[NO_CODE]: The skill is composed entirely of natural language instructions in markdown format. It does not include scripts, executable files, or package manifest files that could facilitate code execution or remote downloads.

calm-down