The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill manages authentication tokens and browser sessions stored in .auth/<env>-<role>.json. While it includes instructions to avoid printing credentials and to read them only by name, the handling of session files represents a sensitive data management surface used for environment access.
[COMMAND_EXECUTION]: The skill makes extensive use of bun run to execute local board commands, sync APIs, and manage sentinels. It also invokes git for repository fetching and drift analysis, and agent launch for worker orchestration.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources including issue trackers (tickets, acceptance criteria), worker reports (mailbox), and board inputs, which could contain malicious instructions.
Ingestion points: Refreshed ticket data from acli/xray-cli, messages from supervised workers via the Run mailbox, and owner instructions from the board inbox.
Boundary markers: The skill uses a structured "7-component briefing" for subagent dispatch and mentions delimiters in the coordinator playbook references.
Capability inventory: Subprocess execution via bun run, repository operations via git, and session management via agent launch and orca-orchestration tools.
Sanitization: The skill implements "read back" verification for ticket statuses and artifact checks (Definition of Done) to validate claims made by workers or external data.

sprint-conductor