The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes complex external data such as feature specifications and user intent ledgers, which constitutes an injection surface.
Ingestion points: Data enters the agent's context from specification files in docs/specifications/, user intent ledgers (user-ledger.md), and terminal screen renderings captured via orca terminal read.
Boundary markers: The skill employs a dedicated screen classification script (scripts/mo-harness-screen.mjs) to verify the state of the harness before delivering task bytes and explicitly instructs the agent not to trust unverified UI or shell prompts.
Capability inventory: The skill can execute Git commands, create GitHub/GitLab issues, and perform extensive filesystem operations within the project worktree.
Sanitization: The instructions include mandatory automated redaction of tokens, passwords, and private keys using the pattern [REDACTED:<kind>] before any implementation data is committed or persisted.
[COMMAND_EXECUTION]: The skill relies on a suite of system tools and bundled scripts to manage the software development lifecycle.
It executes git for version control management, worktree manipulation, and SHA verification.
It utilizes the GitHub (gh) and GitLab (glab) CLI tools for issue tracking and cross-linking methodological friction reports.
It calls the orca orchestration binary to manage interactive agent sessions and verify terminal states.
[DYNAMIC_EXECUTION]: The skill includes a specialized script for environment auditing.
The bundled script scripts/mo-posture.sh performs dynamic shell execution of predefined probe strings to audit the shell environment for alias or function shadowing. This serves as a defensive measure to ensure that legitimate commands (like claude or codex) have not been hijacked by malicious local configurations.

mo-orchestrate-orca