The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it composes agent instructions using data retrieved from local tool discovery. Specifically, it uses 'invocation strings' from installed skills verbatim in the final prompt sent to external agents.
Ingestion points: Data returned by omh coding executor-skills (referenced in references/executor-prompt-composition.md).
Boundary markers: Implements a 'Section Contract' (Goal, Do, Don't, etc.) and an 'Evidence boundary' to delimit sections.
Capability inventory: Executes shell commands via omh coding run and omh coding fanout dispatch, and provides instructions to agents (Claude Code, Codex) that possess file-write and command-execution capabilities.
Sanitization: None observed; the skill explicitly states to use invocation strings 'verbatim' and 'never a guessed prefix'.
[DYNAMIC_EXECUTION]: The skill dynamically assembles a complex prompt ('handoff prompt') from external metadata and command outputs, which is then used to direct the execution of powerful coding agents.
[COMMAND_EXECUTION]: The skill performs multiple subprocess calls to the omh CLI suite and manages the execution of external coding CLIs such as claude-code and codex.
[PRIVILEGE_ESCALATION]: Documentation within the skill suggests the use of the --dangerously-skip-permissions flag when dispatching work to external executors, which bypasses standard permission prompts in non-interactive environments.

ulw-maestro