The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill dynamically spawns and configures worker agents or models at runtime based on external configuration.
references/delegation.md: Describes a 'Dispatch Contract' where the orchestrator spawns agents by name or model ID retrieved from a kickoff.yaml file.
references/delegation.md: Includes logic to discover, select, and store harness-native agent names and model settings, which are subsequently used for execution.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from multiple sources, which could contain malicious instructions.
Ingestion points: Reads engineering briefs, accepted plans, worktree content, and results returned by worker agents (in SKILL.md and references/delegation.md).
Capability inventory: Performs file writes during implementation, spawns worker agents, creates pull requests, and executes the forest CLI tool for closeout.
Boundary markers: Instructions require the orchestrator to 'Inspect every worker diff' and 'Run integrated repository/acceptance validation' to verify outcomes.
Sanitization: Includes instructions to record hashes and names without printing secret contents during closeout.
[COMMAND_EXECUTION]: The skill invokes external command-line tools to manage repository state and task completion.
SKILL.md: Specifically invokes forest close, git ls-files, git worktree list, and create-pr to finalize implementation and delivery.

ship-it