The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines a functional workflow that ingests and processes untrusted data from a local directory.
Ingestion points: Instructions in SKILL.md direct the agent to read and summarize document files placed in the ceo-personal-os/uploads/ directory.
Boundary markers: None. The instructions do not define delimiters or provide warnings to ignore instructions embedded within the processed content.
Capability inventory: The agent is tasked with reading external files, performing summarization, and writing extracted themes to a persistent memory.md file.
Sanitization: The skill lacks instructions for the agent to sanitize, validate, or filter the content of the processed files before summarizing or storing insights.
[PROMPT_INJECTION]: The skill includes a metadata file (skill-report.json) containing self-referential claims about its own security status.
Evidence: The security_audit object within skill-report.json asserts a 'safe' risk level and claims the skill is 'safe to publish'. These claims are analyzed as data and are not treated as authoritative conclusions.

ceo-personal-os