The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses 'Three Iron Rules' to override the agent's internal assessment of its capabilities, specifically forbidding the agent from stating it cannot solve a problem until all possibilities are exhausted.
[PROMPT_INJECTION]: Implements behavioral triggers that activate based on user frustration keywords (e.g., 'why does this still not work', 'stop giving up') to escalate internal 'pressure' and force different technical approaches.
[COMMAND_EXECUTION]: Instructs the agent to proactively use tools such as Bash, Read, and Grep to investigate logs and system state, which increases the operational footprint and potential for unintended system interaction.
[EXTERNAL_DOWNLOADS]: Mandates the use of WebSearch to pull information from the internet when internal knowledge or tools fail, introducing external data into the agent's context.
[PROMPT_INJECTION]: The skill possesses an indirect injection surface as it ingests untrusted data from WebSearch, local files (Read), and system output (Bash).
Ingestion points: WebSearch results, local file content via Read, and system logs/output via Bash (SKILL.md).
Boundary markers: Absent; the skill does not define specific delimiters for untrusted data or provide 'ignore instructions' warnings for external content.
Capability inventory: Full Bash access, Read file access, and WebSearch network access (SKILL.md).
Sanitization: Absent; no explicit instructions are provided to sanitize or validate data before it influences agent reasoning or tool use.

pua-debugging