The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external, untrusted sources which could contain malicious instructions.
Ingestion points: SKILL.md specifies workflows for fetching content from arbitrary blog/article URLs and GitHub PR URLs provided by the user.
Boundary markers: The skill lacks explicit instructions to treat the fetched content as data only, increasing the risk that embedded commands in an article could be executed by the agent.
Capability inventory: The skill utilizes the linkedin_create_linkedin_post tool, giving it the ability to perform authenticated actions on the user's behalf via Composio.
Sanitization: There is no evidence of sanitization or filtering of the external content to prevent prompt injection attacks.
[METADATA_POISONING]: There are contradictory instructions between the skill's operational rules and its evaluation suite.
Evidence: The evals/evals.json file expects hashtags in the generated output, whereas the SKILL.md and references/linkedin-format.md files explicitly forbid the use of hashtags as a non-negotiable rule. This inconsistency may lead to confusing behavior or improper operation during automated testing.

linkedin-post-generator