The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill utilizes the dynamic context injection syntax (!command) within the Context section to execute shell commands when the skill is loaded.
Evidence: Found in SKILL.md executing git branch --show-current, git status --porcelain=v2 --branch, and git log --format='%h %s' --max-count=8.
These commands are standard for developer workflows and are used here to provide the agent with immediate situational awareness regarding the current branch and repository state.
[COMMAND_EXECUTION]: The skill is configured with specific allowed tools to execute git operations.
Evidence: Frontmatter allowed-tools includes Bash(git status *), Bash(git log *), and Bash(git diff *).
The execution instructions (Step 1 and Agentic Optimizations table) direct the agent to use these commands to ground the "Done" bucket against the actual state of the disk, ensuring accuracy before refreshing the plan.
[INDIRECT_PROMPT_INJECTION]: The skill contains a surface for indirect prompt injection as it processes untrusted user data and conversation history to influence agent planning.
Ingestion points: The skill accepts an optional free-text focus directive via $ARGUMENTS and scans the current session's conversation history in Step 1.
Boundary markers: The instructions explicitly warn the agent that the directive should be treated as steering rather than an override. It mandates that genuine in-flight boundaries set by the user previously must not be dropped, even if the directive appears to contradict them.
Capability inventory: The skill has access to file system inspection (Grep, Glob, Read), state modification (TodoWrite), and git command execution (Bash).
Sanitization: The instructions implement a verification step where the agent must ground "Done" claims against the git log and git status rather than relying solely on claims made within the conversation history.

refocus