The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill allows the agent to fetch and process data from arbitrary external URLs provided at runtime (e.g., https://github.com/org/repo/instincts.yaml). This facilitates the ingestion of untrusted content into the agent's internal state.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the 'instinct' import process.
Ingestion points: The <file-or-url> parameter used in the /instinct-import command in SKILL.md.
Boundary markers: The skill documentation mentions format validation but does not describe the use of delimiters or 'ignore' instructions to prevent imported content from being interpreted as commands by the LLM.
Capability inventory: The skill utilizes Bash, Read, Write, Grep, and Glob tools, which could be exploited if malicious instincts are imported.
Sanitization: While basic format parsing is mentioned, there is no evidence of sanitization against prompt injection or malicious instructions embedded within the YAML files.
[COMMAND_EXECUTION]: The skill executes a local Python script ~/.claude/skills/continuous-learning-v2/scripts/instinct-cli.py to handle the import logic, which processes external input provided via the command line.

instinct-import