The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external tools and local files to automate workflow steps, creating a surface for indirect instructions to influence agent behavior.
Ingestion points: The skill reads output from wt inspect, git status, git log, and gh pr view (SKILL.md), as well as local retrospective and observation files like wait-observations.jsonl (references/retrospect.md).
Boundary markers: Absent. There are no delimiters or instructions to treat ingested content as untrusted data.
Capability inventory: The skill can execute git and gh commands, perform eval for session management, and write to the local filesystem.
Sanitization: Absent. Content from the codebase and GitHub is interpolated directly into messages and shell commands.
[DYNAMIC_EXECUTION]: The skill utilizes techniques that execute code or binaries determined at runtime.
Evidence: The command eval "$(wt session set coord-<work-slug>)" executes the output of the wt utility directly in the shell context.
Evidence: The instruction to run ./target/debug/wt --version executes a binary from a build directory, which is a form of dynamic local execution.
[COMMAND_EXECUTION]: The skill relies heavily on shell-based interactions to manage the repository lifecycle.
Evidence: Multiple instances of git (merge, fetch, switch, cherry-pick), gh (api, pr view), and wt (inspect, workflow pass, done) commands are used to manipulate state and communicate between agents.

land