The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes potentially dangerous shell commands to perform its primary function of project reconciliation.
Evidence: Uses git worktree remove, git branch -d, and rm -rf for cleanup tasks.
Mitigations: The skill implements strict safety protocols including canonical path resolution, verification that paths are not system roots (e.g., / or home), and checking that paths are within allowed project prefixes. All destructive actions require explicit user confirmation before execution.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data that could theoretically be manipulated by an attacker to influence agent behavior.
Ingestion points: Data is gathered from local git metadata (git branch -vv) and remote GitHub PR information (gh pr list) in SKILL.md.
Boundary markers: The workflow uses a structured phase-based approach to separate discovery from execution, although it lacks explicit textual delimiters for the ingested data.
Capability inventory: Includes capabilities for file system deletion (rm -rf) and branch management.
Sanitization: Employs comprehensive path validation and symlink escape checks to ensure that even if a malicious branch name or path is processed, the resulting actions remain scoped to the intended directories.

reconcile