The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute local Node.js scripts (query-sessions.mjs, dump-session-parts.mjs, scorecard.mjs) and PowerShell commands (Get-CimInstance, Get-NetTCPConnection, taskkill) to analyze system state and manage processes.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from the opencode.db database and local log files, which may contain content generated by other agents or external tools. This creates a surface for indirect prompt injection, although the skill uses structured JSON extraction to process this data, which limits the risk.
Ingestion points: Reads from the session, message, and part tables in opencode.db, and parses local log files in %USERPROFILE%\.local\share\opencode\log.
Boundary markers: None explicitly defined in the prompts, but the scripts perform field-level extraction rather than raw text ingestion.
Capability inventory: Execution of PowerShell commands for process management (taskkill) and Node.js scripts for database querying.
Sanitization: Use of json_extract to isolate specific metadata (status, exit code) from potentially malicious raw data blobs within the database.
[SAFE]: The skill focuses on read-only diagnostics of the local environment. It includes instructions to avoid unnecessary process termination and uses built-in Node.js modules for all database operations without external network dependencies.

opencode-subagent-log-triage