The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The file scripts/evolve-soul.md contains a hardcoded database password (36795379) in a sample connection string provided in the 'How to run SQL with psql (Windows)' section.
[COMMAND_EXECUTION]: The Python scripts scripts/evolution_proposal.py and scripts/insight_bundle_b012.py use the importlib library to dynamically load and execute code from a local path (specifically ../../skill-system-memory/scripts/mem.py). This dynamic loading of executable content from computed paths increases the risk of local command execution if the file system is compromised.
[PROMPT_INJECTION]: The skill implements an 'Observe and Evolve' loop that processes session transcripts to update the agent's behavior and personality profile, creating a surface for indirect prompt injection.
Ingestion points: Raw session transcripts are processed in scripts/extract-facets.md and prompts/facet-extraction.md.
Boundary markers: The prompts used for extraction (e.g., prompts/facet-extraction.md) do not use delimiters or instructions to ignore instructions embedded within the processed transcript.
Capability inventory: The skill has the capability to write to the filesystem (fs.write) to update agent profiles and workflow recipes, and to the database (db.write) to store state.
Sanitization: There is no evidence of sanitization or filtering of the transcript content to prevent malicious instructions in the data from influencing the agent's long-term behavioral configuration.

skill-system-insight