Step 1 of 4
Decompose the current prompt
Name every instruction's job before deciding where it lives.
Take my current system prompt and decompose it. Classify every sentence into: identity (who the agent is), capabilities (what it can do), hard rules (must/never statements), tool guidance (when to use which tool), environment (paths, stack, versions), style (tone, format), and dead weight (contradictory, duplicated, or unverifiable lines). Produce a table of sentences to categories, flag duplicates and contradictions, and list the dead weight for deletion. Report the token count of each category.
Expected after this step
A categorized inventory of the existing prompt with token counts.
Should not happen
- ✕One monolithic prompt where nobody knows which line does what
- ✕Stale baked-in facts (dates, paths) contradicting reality at runtime
- ✕Rules stated in the prompt but violated freely because nothing enforces them
- ✕Prompt bloat: sections kept out of superstition rather than measured value
Verify before continuing
Do not move on until every check is true. The complete button stays locked until then.
Do not continue if…
- !One monolithic prompt where nobody knows which line does what
- !Stale baked-in facts (dates, paths) contradicting reality at runtime
- !Rules stated in the prompt but violated freely because nothing enforces them
- !Prompt bloat: sections kept out of superstition rather than measured value
If the AI messes this up
Use this when the AI fakes progress or breaks the feature. It forces a real fix.
Everything got labeled 'rules'. Re-classify with the test: identity shapes persona, rules constrain actions, guidance informs choices - if violating the line should stop the run, it is a hard rule; otherwise it is not.