Personal Operating Manual¶

The repeatable system you trust. Read once, then operate by reflex.

The default day¶

1. Open a fresh chat in Cursor on the project you're working in.
2. State the task in 1-2 sentences. Mention scope and data classification.
3. The agent uses the default task loop (ask → plan → bounded edit → test → verifier → commit).
4. Update instrumentation/daily-log.md on every task.
5. End the day with one line in the log noting friction events and any tool-block events.

The week¶

• Monday-Thursday: default loop, default loop, default loop.
• Friday: weekly retro using weekly-retro-template.md. Pick one calibration change for next week.

Triggers¶

Capabilities I commit to¶

• Bounded edit loop is non-negotiable. No "while we're here" cleanups.
• Verifier subagent runs on every task before commit.
• Skills get added via /skill-factory, never hand-coded ad-hoc.
• New projects pull rules via /rules-bootstrap.
• Sensitive captures go to AnyType, not chat history, not the wiki.
• One new capability per week. Calibration is one variable per week.

Capabilities I refuse¶

• Multi-file refactors disguised as small fixes.
• Skipping the verifier "because it's small".
• Manual rule duplication across projects.
• Storing user-level credentials anywhere agents can read them.
• Auto-running shell or MCP outside the project allowlist.

Bug-out plan¶

• If verification keeps failing → return to plan mode and re-scope.
• If a capability is hurting more than helping → kill it for one week, re-evaluate.
• If something feels off → trust the feeling; pause and check the loop.

Reference¶

• Workspace README
• Default task loop
• Security baseline
• Handoff readiness rubric