The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses common system utilities to manage ticket data and repository structure.
Evidence: scripts/sift-gate.sh and assets/scripts/sift.sh utilize git, find, sed, awk, and mkdir to interact with the file system. These operations are scoped to the project root resolved by sift-gate.sh.
[DYNAMIC_EXECUTION]: The initialization process generates a shell script hook for project environment setup.
Evidence: scripts/sift-init.sh creates the config/worktree-setup script using inputs provided through --worktree-link and --worktree-command. The skill implements security controls by validating commands with bash -n and restricting symlink paths to relative locations within the project.
[INDIRECT_PROMPT_INJECTION]: The skill processes ticket data from files, creating an attack surface for untrusted repository content.
Ingestion points: Ticket files (.md files with YAML front-matter) located in .ai/sift/open/ and .ai/sift/archive/.
Boundary markers: The skill relies on standard YAML delimiters (---) and Markdown headers.
Capability inventory: The sift.sh script performs file reading, search, moving, and archiving (mv, find, grep).
Sanitization: Input validation is performed on prefixes and milestones. Data extraction uses awk and sed patterns rather than dynamic execution of content.

sift-init