opencode-conversation-analysis

This skill's stated purpose (analyzing OpenCode conversation histories for themes in user steering behavior) aligns with its capabilities: it locates sessions, extracts user messages, chunks them, runs parallel subagents to analyze chunks, and synthesizes results. The primary security concerns are data exposure and exfiltration of sensitive conversation content. Specifically: (1) the pipeline writes raw conversation chunks to /tmp and dispatches them to multiple subagents — increasing the number of places sensitive data is exposed; (2) subagents are asked to return direct quotes verbatim, which can include secrets or PII; (3) there is no described sanitization, access control, or secure cleanup of temporary files; (4) the use of parallel subagents creates a transitive trust surface (each subagent must be trusted). There are no obvious external downloads or curl|bash install-execute patterns in the provided text, so classic supply-chain download-execute risks are low. Overall, this skill should be considered functionally useful but moderately risky in environments with sensitive conversation data unless mitigations are added: sanitize quotes, redact secrets before chunking, restrict subagent runtime permissions, encrypt/remove temporary chunks, and audit/confirm subagent code. Recommended action: treat as SUSPICIOUS for sensitive data environments and apply the mitigation steps above before use.