The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python helper script, scripts/wingman_store.py, to manage the agent's state. This script is used to initialize directories and perform CRUD operations on Markdown-based profiles and logs located in the user's local state directory.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data via chat screenshots provided by the user for transcript extraction and coaching.
Ingestion points: Chat screenshots are ingested and transcribed according to the instructions in SKILL.md and references/ocr-extraction.md.
Boundary markers: The instructions do not specify explicit delimiters to isolate untrusted text, but they do mandate a human-in-the-loop 'Confirmation Step' where the user must verify the transcript before the agent proceeds with analysis.
Capability inventory: The skill possesses file read/write capabilities through the scripts/wingman_store.py utility script, which manages local storage.
Sanitization: The primary sanitization mechanism is the required user confirmation of the OCR transcript, which serves to prevent the agent from following any malicious instructions that might be embedded in the images.
[EXTERNAL_DOWNLOADS]: The readme.md file contains an installation command using npx that targets the author's repository. This is the standard deployment mechanism for the skill and is documented as a neutral finding regarding the vendor's resources.

crushable-wingman