The Agent Skills Directory

[SAFE]: The skill defines a structured 'plan-execute-verify' loop focused on persistent goal achievement. It incorporates essential security boundaries: single-step retry limits (max 5), mandatory user pauses for financial or irreversible operations (e.g., deletions, publishing), and time-based status reporting (every 30 minutes). These constraints effectively mitigate risks associated with autonomous loops and resource exhaustion.
[PROMPT_INJECTION]: While the skill uses strong directives to encourage persistence ('do not stop until the goal is achieved'), these are task-oriented and balanced by safety guidelines. No attempts to bypass core AI safety filters or override system-level instructions were detected.
[EXTERNAL_DOWNLOADS]: The skill references the use of 'web_fetch' to search for technical solutions on platforms like GitHub or forums during troubleshooting. This behavior is informative/retrieval-based and does not involve the execution of unverified remote code or scripts.

ralph