recall
Pass
Audited by Gen Agent Trust Hub on Nov 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The script
scripts/history.pyexecutesgitcommands using thesubprocessmodule to resolve repository metadata. - Evidence: It uses
subprocess.runto call commands likegit worktree listandgit rev-parseon the target repository path. - Security Note: These calls are implemented using argument lists rather than shell strings, which is a secure practice that prevents shell-based command injection.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from previous chat sessions and shared records, which is a known attack surface for indirect prompt injection.
- Ingestion Points: Historical transcript data is read from local directories (
~/.codex,~/.claude) and external sources (issue trackers, PRs) via the agent's MCP tools. - Boundary Markers: The skill documentation in
references/history.mdexplicitly states that history output is marked asuntrustedEvidenceto alert the agent to potential risks. - Capability Inventory: The skill utilizes the
gitandghcommand-line tools and has the ability to spawn subagents using the/swarmrole. - Sanitization: The
SKILL.mdinstructions include a mandatory step to sanitize private context before any public output is generated.
Audit Metadata