The Agent Skills Directory

[SAFE]: The skill implements strong access control by explicitly defining the required and writable state paths in the YAML frontmatter, limiting the agent's reach to specific strategy and retrospective directories.
[SAFE]: A human-in-the-loop (HITL) mechanism is enforced. The workflow requires the agent to present a draft and wait for an explicit 'APPROVE' command before executing any write operations to the primary strategy/okrs.md or strategy/q-plan.md files.
[SAFE]: Secrets management is handled correctly via environment variable requirements (PEPPER_CLOUD_URL, PEPPER_API_KEY). No sensitive credentials or API keys are hardcoded in the instructions or scripts.
[SAFE]: The use of shell commands is restricted to date manipulation and internal state helpers (state_read, state_write), with no evidence of arbitrary command execution or network exfiltration to external domains.

quarterly-planning