The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from markdown plan files to drive its orchestration logic and generate prompts for subagents.
Ingestion points: The $path argument points to a plan file that is read and analyzed in Step 2 of the reference files.
Boundary markers: The instructions do not include specific delimiters or warnings to subagents to ignore instructions that might be embedded within the plan's work items.
Capability inventory: The system can perform file system modifications, execute shell commands (git, build scripts), and access the network via git. Subagents created by the orchestrator are granted similar powers.
Sanitization: The skill lacks logic to sanitize or escape plan content before it is included in subagent objectives or instructions.
[COMMAND_EXECUTION]: The skill executes various shell commands for version control and project verification.
Evidence: Systematic use of git commands including git branch, git log, git fetch, git rebase, git status, git add, and git commit across references/claude-code.md, references/copilot-cli.md, and references/generic.md.
Evidence: Instructions to run arbitrary build and testing commands as defined by the project environment.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to keep the local repository in sync with the remote origin.
Evidence: Execution of git fetch origin main and git rebase origin/main to ensure the work is based on the latest remote state.

implement