copilot-history-ingest
Pass
Audited by Gen Agent Trust Hub on May 7, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes local system tools such as
lsandsqlite3to discover and query the Copilot session metadata and global history database. These operations are restricted to the user's local environment and are essential for identifying the data to be processed.\n- [PROMPT_INJECTION]: The skill processes potentially untrusted content from GitHub Copilot conversation logs and VS Code transcripts to generate wiki pages (Indirect Prompt Injection surface).\n - Ingestion points: The skill reads from
~/.copilot/session-store.db,events.jsonl, and VS Code transcript files.\n - Boundary markers: Page content is structured using YAML frontmatter, though the synthesized body text does not use specific delimiters to separate source data from generated content.\n
- Capability inventory: The agent is permitted to create and update markdown files and a JSON manifest within the designated Obsidian vault directory.\n
- Sanitization: The instructions explicitly direct the agent to distill and synthesize knowledge rather than performing verbatim copies, and to filter out secrets, API keys, and other sensitive credentials discovered in the logs.
Audit Metadata