The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes user-defined specifications and existing skill files to generate or modify executable code. This surface for indirect prompt injection is mitigated by the inclusion of a mandatory 'adversarial-reviewer' agent (using Claude Opus 4.6) which analyzes implementation plans for failure modes and safety concerns before any changes are applied.
[COMMAND_EXECUTION]: The workflow relies on the Bash tool to execute session management scripts, perform git operations (commit, stash, reset), and run configuration synchronization tools (sync-config.py). These commands are essential to the skill's function as a configuration editor and follow standard development practices.
[SAFE]: No malicious patterns, obfuscation, or unauthorized data exfiltration were detected. Hardcoded absolute paths found in shell scripts (e.g., in 'experimental-tagging.sh') refer to the author's local development environment and do not represent a security threat. The skill demonstrates defensive design through its use of isolated agent contexts and strict pre-flight safety checks.

skill-editor