The Agent Skills Directory

[SAFE]: No security issues were detected. The skill operates on local project files using standard system utilities and follows a structured validation workflow for all modifications. All external vendor resources associated with the skill author are consistent with normal functionality.
[INDIRECT_PROMPT_INJECTION]: The skill processes "delta documents" and implementation reports from thread folders to update project-wide documentation and agent instructions, creating a surface for indirect injection.
Ingestion points: The delta/ directory and report.md files within the target thread folder as defined in SKILL.md.
Boundary markers: Edits require literal matching of existing text and verification of the target file's git blob hash according to references/formats/delta-document.md.
Capability inventory: The skill can create, edit, or delete files in docs/adr/, docs/pdr/, docs/glossary.md, and project agents files (AGENTS.md/CLAUDE.md). It also uses shell utilities including git, wc, grep, and printf for data processing and metadata updates.
Sanitization: Implements a 1,000-word budget for agents files and a requirement to verify that passages remain semantically true after landing, as described in SKILL.md.
[PROMPT_INJECTION]: The skill contains instructions for the agent to proceed with file modifications without additional user confirmation once preliminary checks are successful.
Evidence: "Once the checks pass, the writes run without questions." in SKILL.md.
[COMMAND_EXECUTION]: The skill utilizes local shell commands for repository introspection, file validation, and log management.
Evidence: Usage of git ls-files, git hash-object, wc -w, git log, git diff, grep, and printf for file operations and validation throughout the skill body and instructions.

close-thread