The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill performs periodic package support checks by fetching a JSON advisory from the author's official repository on GitHub. This is handled by scripts/core/support/cache.py using Python's urllib.request.urlopen. The URL targeted is https://raw.githubusercontent.com/Frappucc1no/recall-loom/main/release-advisory.json. This behavior is documented in the skill's support policy and is used solely for maintenance and versioning.
[COMMAND_EXECUTION]: The skill utilizes subprocess.run in several scripts for legitimate coordination and discovery tasks:
scripts/recallloom.py uses it as a unified entry point to execute other internal Python scripts based on user commands (e.g., init, validate).
scripts/core/coldstart/sources.py uses it to execute git log to extract recent commit history as an optional signal for project context.
[DATA_EXFILTRATION]: The skill implements a dedicated defensive layer in scripts/core/safety/attached_text.py. This script contains regex patterns to scan text for sensitive content (tokens, API keys, SSH keys) and prompt injection markers (e.g., "ignore previous instructions"). This is a security feature designed to prevent the agent from accidentally reading or writing malicious or sensitive content during context restoration.
[NO_CODE]: The skill provides templates in the native_commands/ directory for host-specific integrations (Claude Code, Gemini CLI, OpenCode). These templates utilize platform-native execution syntax (e.g., ! commands) to call the RecallLoom dispatcher. These are intended to be installed locally by the user to provide convenient shortcuts.

recallloom