The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses various shell commands (duoduo daemon, duoduo channel, npm, bash) to manage the local installation. These commands are consistent with the skill's primary purpose of host-mode administration.
[EXTERNAL_DOWNLOADS]: The skill facilitates software updates through npm and fetches instruction updates from the vendor's GitHub repository (github.com/openduo/duoduo). These are trusted, well-known services and organizations.
[INDIRECT_PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by ingesting local configuration files (like .env and descriptor.md) and command outputs into the agent context. However, the risk is mitigated by explicit instructions to avoid leaking private information and to restate symptoms clearly before acting. Mandatory evidence chain:
Ingestion points: scripts/v05-upgrade-preflight.sh, duoduo daemon config, and local .env file reads.
Boundary markers: None explicitly used in prompts.
Capability inventory: File system access, network access (via npm/skills CLI), and daemon process control (duoduo daemon restart).
Sanitization: Not present; however, the skill mandates user verification and warns against silent actions.
[PROMPT_INJECTION]: Static detectors flagged potential concealment attempts; however, manual review confirms these are actually defensive instructions. The skill repeatedly emphasizes 'Offer — do not do it silently' and 'Do not silently rewrite files', which are anti-concealment measures designed to maintain user oversight.

duoduo-admin