migrate-copilot-instructions
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process instruction files from a repository (such as
.github/copilot-instructions.md) and port their content into new configuration files for other AI agents. This creates a surface where malicious instructions embedded in the source repository files could influence the behavior of the target agent after migration.\n - Ingestion points: Configuration and instruction files located in
.github/,.claude/,.codex/, and.opencode/directories.\n - Boundary markers: The instructions emphasize a "convert, don't copy" approach and advise stripping tool-specific language, but do not mandate explicit boundary markers or "ignore embedded instructions" warnings in the generated output.\n
- Capability inventory: The skill performs file reading, file writing, and shell command execution (e.g., the detection script in Step 0).\n
- Sanitization: The skill recommends normalization and re-interpretation of instructions rather than raw copying to reduce the risk of inheriting incompatible or malicious directives.\n- [COMMAND_EXECUTION]: The skill uses a shell script (Step 0 in
SKILL.md) to detect the presence of various agent-related configuration files in the local repository. This is a legitimate functional requirement for the migration process and does not involve untrusted input injection.
Audit Metadata