The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill downloads a shell script ('install.sh') from 'https://raw.githubusercontent.com/Necmttn/ax/main/install.sh' and executes it using bash. The instructions include a manual review step using 'less' and require the user to confirm execution, which provides a layer of oversight for this remote code execution pattern.
[EXTERNAL_DOWNLOADS]: The command 'npx skills add Necmttn/ax' is used to download and register additional agent skills from a remote repository into the local environment.
[PERSISTENCE]: The skill documentation mentions the installation of a macOS LaunchAgent ('com.necmttn.ax-otlpd') for optional telemetry. LaunchAgents are a standard mechanism for ensuring processes run automatically at user login.
[INDIRECT_PROMPT_INJECTION]: The 'ax ingest' command reads local Claude Code and Codex transcripts to build its graph. This creates an attack surface where instructions embedded in processed data could influence agent behavior.
Ingestion points: Reads transcript files from the directory specified by 'AX_TRANSCRIPTS_DIR'.
Boundary markers: No specific delimitation or 'ignore instructions' markers are mentioned for the ingested transcript data.
Capability inventory: The skill can execute shell commands ('bash', 'ax', 'npx', 'curl'), write to the filesystem, and perform network operations.
Sanitization: The skill does not describe any validation or filtering performed on the content of ingested transcripts.

setup