The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves and summarizes past session history from the OpenCode API and local database. This data is untrusted as it contains past user and assistant messages which may contain instructions that could influence the agent's behavior during the summarization process.
Ingestion points: references/api.md specifies reading messages via GET /session/{id}/message and the local SQLite database at ~/.local/share/opencode/opencode.db.
Boundary markers: The skill instructs the agent to "Exclude secrets and sensitive transcript data," but lacks explicit delimiters to isolate untrusted session history from the agent's primary instructions.
Capability inventory: The skill utilizes curl for network requests to a local server, sqlite3 for database queries, and git for repository analysis.
Sanitization: Instructions require the exclusion of secrets but do not provide specific guidance on sanitizing or escaping the content of retrieved messages.

summarize-work