The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on several embedded Bash scripts to manage the orchestration workflow. These scripts perform local tasks such as updating the <plan>.orchestration-status.md file, checking pre-flight requirements in <plan>.preflight.md, and aggregating audit data into <plan>.audit-pack.md. These operations use standard text-processing utilities like awk and cat and are restricted to the local filesystem.
[PROMPT_INJECTION]: As a workflow orchestrator, the skill ingests a user-provided specification document ($ARGUMENTS) which is passed through a chain of sub-agents (e.g., review-design, mece-plan-review). This creates an indirect prompt injection surface where a malicious specification could attempt to influence the behavior of the sub-agents. However, the skill implements 'machine gates' and human-in-the-loop checkpoints to mitigate automated execution risks.
[DATA_EXFILTRATION]: The skill aggregates project metadata, design reviews, and quality ledgers into an 'Audit Pack' file. The analysis confirms that this information is stored locally for the user's review and is not transmitted to any external domains or third-party services.
[REMOTE_CODE_EXECUTION]: While the skill mentions version checking by comparing local files against potential plugin installations, it does not download or execute code from remote servers. All skill invocations are performed by reading local SKILL.md files from established plugin or configuration directories.

deliver-from-spec