cc-new-feature
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill contains a Sovereignty Declaration that explicitly instructs the agent to ignore or prohibit the use of other tools during its workflow.
- Evidence: Found in SKILL.md: 禁止在本流程执行期间调用 superpowers:brainstorming... 或任何会接管设计/实现流程的外部 skill.
- [COMMAND_EXECUTION]: The skill directs the agent to execute shell commands to manage its directory structure.
- Evidence: mkdir -p .codex/tasks/archived specified in SKILL.md and agents/openai.yaml.
- [PERSISTENCE]: The skill maintains task state across sessions by writing data to the file system in the project root or the user home directory.
- Evidence: Storage logic for task persistence described in SKILL.md and references/resume-policy.md.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user requirements and writes them into task files which are later read to resume work, creating a surface for injection.
- Ingestion points: User feature requirements (SKILL.md).
- Boundary markers: No delimiters are defined in templates/task.md to separate data from instructions.
- Capability inventory: File creation, writing, and shell command execution.
- Sanitization: No input validation or escaping is performed before storage.
Audit Metadata