The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions include several shell command blocks using find, grep, ls, sed, and awk for repository auditing and file discovery. These commands are executed to identify files matching specific patterns, such as all-caps root files or files referenced in other documentation.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes content from various repository files, including README.md, AGENTS.md, documentation files, package.json, and .context/ artifacts, to update other documents. This creates an attack surface where instructions embedded in project data could influence the agent during the synchronization process.
Ingestion points: references/sync.md specifies reading project identity files, technical context files, and live configuration like package.json to gather data for synchronization.
Boundary markers: The skill defines specific sections to preserve verbatim in critical files like AGENTS.md and README.md to prevent overwriting structural components.
Capability inventory: The skill is designed to perform extensive file reading and writing operations across the repository.
Sanitization: A dedicated security protocol is implemented in references/sync.md to scan for and redact credentials (passwords, tokens, API keys) and production URLs before writing files to disk.

sync-ai-context