The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's 'Execution Log' instruction specifies a shell command (echo ... >> ...) that appends a summary of input (8_WORD_SUMMARY) to a local file. This summary is generated by the agent based on user input and is not sanitized for shell-active characters like single quotes, which could lead to arbitrary command execution.
[COMMAND_EXECUTION]: In the extraction workflow, the skill executes the fabric CLI tool using the pattern fabric -y "URL". Without strict validation or escaping of the user-provided URL, this pattern allows for shell command injection.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources, which can contain malicious instructions.
Ingestion points: YouTube transcripts retrieved via fabric and article content retrieved via WebFetch.
Boundary markers: No specific delimiters or 'ignore' instructions are used to isolate ingested content from the agent's core instructions.
Capability inventory: The skill possesses the ability to execute shell commands (fabric, echo) and read/write to the user's home directory (~/.claude/).
Sanitization: No sanitization or filtering of external content is performed before processing.
[DATA_EXPOSURE]: The skill interacts with the local file system by reading user-specific customization files from ~/.claude/LIFEOS/USER/CUSTOMIZATIONS/ and writing to an execution log in ~/.claude/LIFEOS/MEMORY/SKILLS/execution.jsonl.
[EXTERNAL_DOWNLOADS]: The skill retrieves content from external URLs (YouTube and articles) using the fabric tool and WebFetch. While these are vendor-integrated tools, the lack of sanitization on the inputs provided to them is a security concern.

ExtractWisdom