The Agent Skills Directory

[COMMAND_EXECUTION]: The skill exposes functionality for executing arbitrary shell commands through a 'terminal dock' feature, controlled via the terminal_exec backend command. This capability allows the agent to perform a wide range of system operations within the context of a workspace.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of data from external sources, such as Git repositories and GitHub (via the gh CLI), which can be influenced by third parties.
Ingestion points: Untrusted data is retrieved using commands like git_diff_stats, github_list_issues, github_list_prs, and github_pr_diff within SKILL.md and referenced core logic.
Boundary markers: The instructions do not provide explicit delimiters or guidance to ignore instructions embedded within the ingested data, which could lead to the agent inadvertently following malicious commands found in pull requests or commit messages.
Capability inventory: The agent possesses significant capabilities, including shell access (terminal_exec), file system manipulation (worktree_create), and network interaction via the daemon mode.
Sanitization: No sanitization or validation logic for the external content is described in the skill's instructions or core logic snippets.
[EXTERNAL_DOWNLOADS]: The documentation references external installation requirements including the Node.js ecosystem, the Rust toolchain, and native build tools like CMake and Clang.

codexmonitor-orchestration