The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill uses sops to decrypt production secrets and extracts sensitive tokens like POLAR_ACCESS_TOKEN using shell utilities (grep, cut). Evidence found in SKILL.md and scripts/update-tier.sh.
[COMMAND_EXECUTION]: Multiple scripts execute shell commands via npx wrangler d1 to directly run SQL against a production database (DB). This includes UPDATE and SELECT operations based on user input. Evidence in scripts/update-tier.sh and scripts/sync-d1-polar-balances.ts.
[DATA_EXFILTRATION]: The skill facilitates data movement by fetching user data from a production database and sending it to external services (GitHub APIs for commit/stargazer checks and Polar for billing). While these are functional requirements, the broad network access combined with database read access is a high-capability pattern.
[PROMPT_INJECTION]: The SKILL.md includes 'IMPORTANT' markers instructing the agent to ensure dual updates to both the database and subscription services, which attempts to override default decision-making logic for critical billing paths.

tier-management