The Agent Skills Directory

[PERSISTENCE]: The skill provides a hook install command that modifies the .claude/settings.json file to execute a dashboard command at the start of every agent session. While this is a documented feature for the 'home' view dashboard, modifying agent configuration files is a persistence mechanism.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user feedback and comments via the feedback export command for LLM processing. This data could contain instructions meant to influence the agent's behavior.
Ingestion points: The feedback export command in scripts/drafter-axi.mjs fetches submission data from the configured external API.
Boundary markers: The skill does not provide explicit prompt delimiters or instructions for the agent to ignore embedded commands in the exported data.
Capability inventory: The skill has network access via fetch and file system write capabilities via node:fs.
Sanitization: No security-focused sanitization is performed on the ingested feedback before it is presented to the LLM.
[COMMAND_EXECUTION]: The script scripts/drafter-axi.mjs uses node:child_process.execSync to run git rev-parse --show-toplevel for project directory discovery during the hook installation process.
[CREDENTIALS_UNSAFE]: The skill handles sensitive administrative API tokens (DRAFTER_ADMIN_TOKEN) and includes commands like people links that generate and output participant sign-in tokens.

drafter-axi