oceinteg

Pass

Audited by Gen Agent Trust Hub on Oct 6, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructions direct the agent to execute shell commands for managing credentials, environment configuration, and repository state. Specific commands mentioned include op (1Password CLI) for secret retrieval, jq for JSON parsing, and system utilities like umask and install for setting directory permissions. It also requires the use of git for repository operations including cloning, pushing, and deleting branches.
  • Evidence: op item get "$ITEM_NAME" --vault "$VAULT_NAME" --format json | jq ... and op read "$GITHUB_APP_KEY_REF" --file-mode 0600 --out-file ... in references/credentials.md. git clone, git fetch, and git push operations described in references/main.md and references/runtime-acceptance.md.
  • [INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface by ingesting and acting upon instructions from workspace files and external communication channels during its execution.
  • Ingestion points: The agent is instructed to read instructions from a file named AGENTS.md within the workspace and to process messages from Slack channels (references/main.md).
  • Boundary markers: The skill includes a protocol to only follow instructions in AGENTS.md that name specific authorized individuals ("Kevin or Peter") and match a unique run ID. However, this relies on the agent's interpretation of the text rather than hard enforcement.
  • Capability inventory: The agent has access to highly sensitive capabilities, including retrieval of Slack and GitHub credentials via op, deployment of infrastructure via Helm/Kubernetes, and the ability to push code to repositories.
  • Sanitization: There is no mention of programmatic sanitization or filtering of the instructions received from AGENTS.md or Slack.
  • [EXTERNAL_DOWNLOADS]: The skill involves downloading and deploying external assets as part of the integration testing workflow.
  • Evidence: Instructions to choose and pull "published release or operator-supplied custom immutable controller/runtime/broker images" during setup (references/main.md).
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 6, 2026, 11:38 AM