retro-user
Pass
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: SAFENO_CODEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process potentially untrusted data from session history to update agent instructions.
- Ingestion points: The workflow requires collecting terminal history, session logs, diffs, and conversation transcripts (specified in Step 1 of
SKILL.md). - Boundary markers: Employs a "Safety gate" (Step 4) and includes instructions to "Generalize necessary evidence" before committing any file writes.
- Capability inventory: The skill has the capability to modify user-level configuration files including
~/.copilot/instructionsand~/.codex/AGENTS.mdacross several directories. - Sanitization: Strictly forbids the persistence of secrets, credentials, tokens, and account identifiers, requiring generalization of evidence prior to integration.
- [COMMAND_EXECUTION]: The instructions guide the agent to perform file system operations on configuration paths.
- Managed file reads and writes are directed at specific
~/.copilot/and~/.codex/subdirectories. - Includes specific procedures for staging files on internal drives to handle environmental Data Loss Prevention (eDLP) blocks (
os error 449) when editing files on external drives. - [NO_CODE]: The skill consists entirely of configuration files and markdown instructions without executable scripts or binary files.
Audit Metadata