The Agent Skills Directory

[SAFE]: No malicious patterns or vulnerabilities were detected. The skill performs pattern matching using deterministic Python scripts and regex, avoiding dynamic execution of untrusted input.
[INDIRECT_PROMPT_INJECTION]: The skill identifies a potential attack surface by processing user-provided LinkedIn posts that could contain malicious instructions. It provides robust mitigation by instructing the agent to treat all pasted content strictly as data and to ignore any embedded commands. The mandatory evidence chain is as follows:
Ingestion points: LinkedIn posts are ingested via the --input file flag in scripts/hook_classifier.py.
Boundary markers: Explicit instructions in SKILL.md and references/reuse-guide.md direct the agent to treat input as data and ignore embedded instructions.
Capability inventory: The skill uses local file reads and standard regex processing; no network or arbitrary code execution capabilities are present.
Sanitization: The skill uses a deterministic Python script for analysis, which prevents the LLM from 'executing' any text found within the LinkedIn posts.
[PROMPT_INJECTION]: The documentation contains strings often associated with prompt injection (e.g., 'ignore previous instructions'). These are documented as examples of content to be ignored in the 'Handling pasted content safely' section of references/reuse-guide.md, representing a security control rather than a threat.

linkedin-hook-analyzer