The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple shell commands to audit the repository and discover AI-critical documents. Evidence includes the use of find, grep, ls, sed, and awk within references/sync.md to scan file contents and directory structures.
[EXTERNAL_DOWNLOADS]: The skill references an external utility, caveman-compress, and provides a link to its repository (https://github.com/JuliusBrussee/caveman). It encourages the execution of this tool if installed at the user level to compress documentation prose.
[PROMPT_INJECTION]: The skill accepts user-supplied $ARGUMENTS and passes them through to the synchronization logic. While the instructions specify these arguments define the target repository, unvalidated input flowing into shell contexts presents a potential injection surface.
[DATA_EXPOSURE]: The skill includes a defined 'Security protocol' designed to prevent the exposure of sensitive data. It explicitly forbids reading .env files and implements a pre-write redaction step to replace detected API keys, tokens, and passwords with placeholders.
[INDIRECT_PROMPT_INJECTION]: The skill processes external repository data to update canonical instruction files.
Ingestion points: Reads package.json, .context/ data maps, root markdown files, and HTML documentation.
Boundary markers: Employs a 'patch-in-place' procedure with specific 'preserve-lists' for critical sections in AGENTS.md and README.md to prevent total document overwrite.
Capability inventory: Performs shell-based file auditing (find, grep) and file-writing operations across the repository.
Sanitization: Implements a redaction log that must be surfaced to the user before any file is written to disk, allowing for manual verification of removed content.

sync-ai-context