The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary workflow involves ingesting external 'Authority' documents (such as design specs, PRDs, or remote URLs) to define the goals of its autonomous loop. This ingestion creates a surface where instructions embedded within those documents could attempt to influence the agent's actions.
Ingestion points: The skill reads user-defined 'Authority' files or URLs during the assessment and delegation phases of every execution round, as specified in references/execution-guide.md.
Boundary markers: Instructions require the agent to use structured state files (FOCUS.md, TASKS.md) and a 'Placement' framework to limit subagent authority, but the content of the external authority source itself is not programmatically sanitized.
Capability inventory: The skill is designed to orchestrate subagents that have the capability to read/write files and execute development commands (e.g., tests and builds).
Sanitization: The protocol relies on a structured human alignment phase to review the generated 'plan' and requires all changes to be verified by external signals (like test suites) rather than the agent's own assessment.
[COMMAND_EXECUTION]: The execution protocol requires subagents to verify their work using external signals such as npm test, build, lint, or type-check. This involves the automated execution of shell commands within the project environment to confirm convergence and correctness.
[EXTERNAL_DOWNLOADS]: The README.md provides installation instructions that fetch the skill from a GitHub repository (github.com/synapseradio/ai-skills) using npx. This repository is associated with the skill's author.

runbook