The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to execute shell commands including npx, git, and local build tools (typecheck, test, build) to process and verify repository changes.
[EXTERNAL_DOWNLOADS]: The skill utilizes the @decocms/blocks-cli package via npx. This package resides within the vendor namespace of the skill's author (decocms) and is a legitimate dependency for the intended reconciliation functionality.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data, specifically git patches and commit logs from an upstream repository. This represents a theoretical attack surface where malicious code or instructions could be introduced via the code being reconciled.
Ingestion points: Patch files (patches/*.patch) and git logs generated by the reconciliation script (stored in manifest.json).
Boundary markers: The instructions provide a logical loop for processing files but do not specify strict prompt delimiters or instructions for subagents to ignore embedded prompts in the patches.
Capability inventory: The agent is authorized to modify local files, execute vendor-provided CLI tools, and run standard repository verification commands.
Sanitization: No specific sanitization or filtering of patch content is performed before the agent translates and applies it to the target repository.

deco-reconcile-snapshot