decodie-observe

Pass

Audited by Gen Agent Trust Hub on Aug 28, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute a bundled shell script (scripts/summarize-index.sh) for indexing and to use shell utilities (shasum, cut) for anchor hashing. These operations occur within the local environment to manage the skill's data directory.
  • [INDIRECT_PROMPT_INJECTION]: The skill's instructions require the agent to load and follow rules defined in .md files within the project's .decodie/rules/ and the user's home directory. This allows potentially untrusted project content to define agent behavior.
  • Ingestion points: Markdown files located in .decodie/rules/ and ~/.decodie/rules/.
  • Boundary markers: None; the skill explicitly directs the agent to treat these files as additional instructions.
  • Capability inventory: Shell command execution (including pipes and script execution) and file system read/write access.
  • Sanitization: The skill does not define any validation or sanitization for the loaded rule files.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 28, 2026, 08:19 PM
Security Audit — agent-trust-hub — decodie-observe