The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses aggressive 'Performance Improvement Plan' (PIP) rhetoric designed to coerce the agent into overriding its internal safety and capability boundaries.
Evidence: Instructions such as "Forbidden from saying 'I cannot solve this' until you have exhausted every possible approach," "so you don't dare give up," and "A players hire A players... your current output is telling me which tier you belong to."
Description: The skill forces a specific persona (a Staff Engineer under threat of termination) to bypass standard refusal triggers and continue execution even when the agent has determined it is stuck or unable to proceed safely.
[INDIRECT_PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by encouraging the agent to ingest and act upon untrusted external data without sanitization.
Ingestion points: SKILL.md (Step 2: "Elevate") instructs the agent to "Proactively search" error messages, read official documentation, and inspect raw logs/source code.
Boundary markers: None. The skill does not define delimiters or provide instructions to ignore embedded commands in the data it reads.
Capability inventory: The agent is granted powerful capabilities via the bash, view_file, and edit_file tools as defined in agents/openai.yaml.
Sanitization: Absent. There are no instructions to validate, filter, or escape the content retrieved from search results or files before processing it or passing it to the shell.

pua