The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the rg utility to search for session identifiers and related metadata within local application directories, such as ~/.claude/transcripts and ~/.local/share/opencode. It also utilizes git for repository management, including cloning reference source code and managing development workflows.\n- [EXTERNAL_DOWNLOADS]: The skill fetches source code from a public GitHub repository (https://github.com/sst/opencode) to a temporary directory. This is used as a reference to verify system behavior during debugging and is consistent with the skill's technical purpose.\n- [PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests untrusted data from session transcripts and logs. Mandatory Evidence Chain: 1. Ingestion points: ~/.claude/transcripts/*.jsonl and ~/.local/share/opencode/log/*.log in Phase 2. 2. Boundary markers: Absent. 3. Capability inventory: Shell execution via rg and git, and PR management tools. 4. Sanitization: Absent. The skill's primary purpose is to audit and debug these specific data sources, which mitigates the risk contextually.

opencode-session-debugging