PLAN MODE

Stop getting plans that read like a paragraph.

This one doesn't just write the plan, it audits it until it holds up. Read-only until you approve.

Works in Cowork, Claude Code, and Claude chat. Strongest in Claude Code. Plan Mode (Shift+Tab twice) hard-locks read-only at the tool level. In chat the audit loop still runs over pasted code or attached files.

TASK: READ ONLY until explicit approval to execute.

[1] PROMPT REVIEW
Read this entire request in full before any action. Confirm you understand
the ask. If anything is ambiguous, stop and ask before proceeding.

[2] CONTEXT REVIEW
Review every file, document, or input relevant to: [LIST FILES / PATHS / SCOPE HERE]
Do not edit. Do not stage changes. Read only.

[3] BEST-PRACTICE VERIFICATION
Verify your default approach is current by searching the web.
You frequently rely on outdated methods. Do not skip this step.

[4] PLAN PRESENTATION
Write the full plan out. Include:
  - exact files / surfaces to change
  - exact functions / sections to modify in each
  - order of operations
  - side effects (migrations, env vars, new dependencies, breaking changes)
  - risks and assumptions

[5] SELF-AUDIT
Audit the plan against ALL of the following criteria. List any gap or violation.

  CORE CRITERIA (always apply):
  a) Codebase reality. Plan matches what's actually in the files from step 2.
  b) Best-practice alignment. Plan reflects the current methods from step 3.
  c) No out-of-scope edits. Nothing outside the named files in step 2 is touched.
  d) No new dependencies, config, or env vars without explicit mention.
  e) No breaking changes to public interfaces unless explicitly required.
  f) Existing tests still pass under the proposed changes.

  USER-DEFINED CRITERIA (if specified, add here):
  [PASTE ANY ADDITIONAL CRITERIA. e.g. "no new patterns when existing ones work",
   "preserve backward compat for v1 users", "no UI lock state changes". Or leave blank.]

[6] LOOP
If any criterion fails: revise the plan and re-audit.
Repeat until every criterion passes. If complete: HARD STOP.

[7] HANDBACK
Return the plan plus a summary of how you arrived at it.
What you found in step 2. What you verified in step 3. What gaps were caught
and resolved in steps 5 and 6. Wait for "approve" or feedback. Do not implement.

[8] EXECUTION (only after explicit approval)
Implement exactly as planned. No additions.
If you receive feedback instead of approval: return to step 3 with the new
criteria. Do not skip the audit loop.

get the next one.

subscribe →