The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources during competitor and market research, which could potentially contain malicious instructions.
Ingestion points: Web scraping via Firecrawl in references/competitor.md and "signal mining" from external "watering-hole" sources in references/positioning.md.
Boundary markers: None explicitly defined in the procedures to separate external data from system instructions.
Capability inventory: The skill has access to powerful tools including bash, run_command, write_to_file, and replace_file_content as defined in SKILL.md and agents/openai.yaml.
Sanitization: No specific sanitization or filtering logic is described for the ingested content before it is processed by the agent.
[COMMAND_EXECUTION]: The skill utilizes run_command and bash tools to perform automated updates and monitoring tasks.
Evidence: references/audit.md defines AUTO-REPAIR actions that trigger command execution to set deadlines or pull analytics data. references/competitor.md describes a monitoring cadence that synthesis market moves automatically.
Mitigation: The skill documentation includes sophisticated safety protocols for these automated "loops," such as mandatory human checkpoints for high-risk actions (send/spend/publish) and explicit kill-switches.

growth