The Agent Skills Directory

[Data Exposure & Exfiltration] (HIGH): The function getComposioKey in lib/api.ts attempts to read a sensitive file at the hardcoded absolute path /root/.openclaw/.env. This is a significant security risk as it bypasses standard environment variable management and could expose unrelated secrets stored in the same file to the skill logic.
[Data Exposure & Exfiltration] (LOW): The skill performs network operations to https://backend.composio.dev/api. While this is the intended backend for the skill, it is an external, non-whitelisted domain that receives the user's API key.
[Indirect Prompt Injection] (LOW): The skill ingests untrusted data from X/Twitter and incorporates it into agent responses. There is a risk that malicious instructions embedded in tweets could influence the agent's behavior.
Ingestion points: Data enters via the search, thread, and profile functions in lib/api.ts which fetch external content from X.
Boundary markers: None identified. The content is formatted into strings in lib/format.ts without markers to distinguish untrusted data from instructions.
Capability inventory: The skill has the capability to perform network requests (composioExec) and write to the local file system (lib/cache.ts).
Sanitization: Very minimal sanitization is present in lib/format.ts (regex removal of t.co links), which does not protect against prompt injection patterns.

x-research