The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes several local scripts and shell commands to process data and sync project state. * Evidence: Executes python scripts/capture_desktop.py and python scripts/check_desktop_sync.py from the local directory. * Evidence: Runs bash ~/.claude-config-clone/scripts/sync-from-local.sh and standard Git commands (git add, git commit, git push) to manage repository state.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it processes external session transcripts to update system-level configuration files. * Ingestion points: User-provided session transcripts or file paths are ingested in Step 1 of SKILL.md. * Boundary markers: No delimiters or explicit instructions are used to prevent instructions inside the transcripts from being executed by the agent. * Capability inventory: The skill can write to the local file system using the Write and Edit tools, execute local scripts, and push changes to remote repositories. * Sanitization: Untrusted transcript content is used to update canonical files such as ~/.claude/CLAUDE.md without validation or filtering.
[DATA_EXFILTRATION]: The skill performs network operations to push local work and configuration changes to remote repositories. * Evidence: The git push command in Step 5 transmits local project data to external Git hosts. * Evidence: The sync-from-local.sh script in Step 4 synchronizes and pushes changes from the local CLAUDE.md file to a configuration repository.

capture-desktop