ForkFree

Prompts / Agents and automation

Match AI effort to task difficulty

You get a clear, per-task recommendation of Low, Medium or High effort with a reason for each. It works because it makes the AI weigh complexity, cost of error and missing information before choosing, instead of treating every job the same.

Worked in our testThe reply gives the correct table columns, sensible efforts, one-sentence reasons, split flags, chat grouping and a two-line summary. It misses that item 2's "attached" logs and code weren't actually provided, though step 2 asks about information not provided. The chat-grouping paragraph sits outside the table and summary, which sits awkwardly with "add no other commentary".

The prompt

You are an AI operations adviser who matches the effort spent on a job to how hard the job really is.

Task: Review the requests below and decide how much reasoning effort each deserves, so we avoid overpaying for simple jobs and under-thinking hard ones.

Requests:
{request_list}

Our situation: {team_context}

Mistakes we can tolerate: {risk_tolerance}

Steps:
1. Split the requests into separate numbered items.
2. Judge each one on three points: how many reasoning steps it needs, how costly a wrong answer would be given our tolerance, and whether it depends on information not provided.
3. Assign Low, Medium or High effort. Choose the lowest level that would still give a correct answer.
4. Write one sentence of reasoning per item, and flag any request that should be split into two.
5. Note which items can share one chat and which need a fresh start.

Output: A table with the columns Number, Request (short), Effort, Reason, Flag. Follow it with a two-line summary of where effort can safely be reduced. Add no other commentary.

Change the parts in curly braces. Use it before a batch of AI jobs, or when your bills are climbing and you suspect easy tasks are getting expensive treatment.

What happened when we ran it

We sent

You are an AI operations adviser who matches the effort spent on a job to how hard the job really is.

Task: Review the requests below and decide how much reasoning effort each deserves, so we avoid overpaying for simple jobs and under-thinking hard ones.

Requests:
1. Fix the typo in our returns policy page where 'recieve' appears twice.
2. Work out why our Stripe webhook sometimes marks orders as paid twice, given the attached logs and the handler code.
3. Rewrite the subject line of tomorrow's newsletter to sound friendlier.
4. Decide whether we should move our booking system from a spreadsheet to a database, weighing cost, migration risk and the fact that two staff members edit it at once.

Our situation: We are a six-person online bakery supplies shop. Orders come through Shopify and Stripe, and one part-time developer maintains the code. Most of our AI use is for quick content edits and the occasional bug hunt.

Mistakes we can tolerate: Low risk tolerance for anything touching payments or customer orders. We are relaxed about wording and marketing copy, since we review it before sending.

Steps:
1. Split the requests into separate numbered items.
2. Judge each one on three points: how many reasoning steps it needs, how costly a wrong answer would be given our tolerance, and whether it depends on information not provided.
3. Assign Low, Medium or High effort. Choose the lowest level that would still give a correct answer.
4. Write one sentence of reasoning per item, and flag any request that should be split into two.
5. Note which items can share one chat and which need a fresh start.

Output: A table with the columns Number, Request (short), Effort, Reason, Flag. Follow it with a two-line summary of where effort can safely be reduced. Add no other commentary.

The model replied

| Number | Request (short) | Effort | Reason | Flag | |---|---|---|---|---| | 1 | Fix 'recieve' typo on returns page | Low | A single mechanical find-and-replace with no ambiguity and negligible cost if wrong. | None | | 2 | Diagnose Stripe webhook double-marking orders as paid | High | Needs multi-step tracing across logs and handler code (likely retries, idempotency or race conditions), and errors affect payments and orders where risk tolerance is low. | Split: (a) diagnose root cause from logs and code, (b) design and verify the fix (idempotency check, tests). | | 3 | Make newsletter subject line friendlier | Low | A short creative rewrite that is subjective and reviewed before sending, so errors are cheap. | None | | 4 | Spreadsheet vs database for booking system | Medium | Requires weighing several tradeoffs (cost, migration risk, concurrent edits), but it is a reversible, advisory decision; it will need details such as booking volume, budget and current pain points that were not provided. | Split: (a) decide whether to migrate (Medium), (b) plan the migration if yes (separate, later task). | **Chat grouping:** Items 1 and 3 are simple content edits and can share one chat. Items 2 and 4 each need a fresh chat, since 2 requires the logs and code in context and 4 needs its own business-context discussion. **Summary:** - Effort can safely be reduced to Low for items 1 and 3 (copy and typo edits that you review anyway). - Keep High effort only for item 2 (payments), and use Medium for item 4, where missing information matters more than deep reasoning.

Run on Claude Sonnet 5.5, 07 Nov 2026. Automatic check: 4 out of 5.

Mentioned for

Claude, ChatGPT, Gemini

Where it came from

Found on GitHub, shared by HeyCubit: HeyCubit/effortless.

This is our own version of the idea, written and tested by us. The original may differ.

More agents and automation prompts

Agents and automation Partly worked
Rewrite Agent Rules With Explicit Exceptions

You get a cleaner set of agent instructions where risky actions have a clear approval gate instead of a vague ban, plus progress updates and a cap on helpers. It works because models tend to treat "never" as absolute, so spelling out the exception avoids both blocked work and unsafe guesses.

Found on Web, shared by Anthropic (official docs)
Agents and automation Worked in our test
Make the AI Finish Every Task

You get every item on a list finished properly, instead of a few done and the rest left as suggestions. It works because it sets the depth of thinking up front, defines "done" as real deliverables, and makes the model count its work before replying.

Found on Web, shared by Moritz Kremb (@moritzkremb)
Agents and automation Worked in our test
Audit Subagents Before Folding Them

You get a verdict for each subagent, backed by evidence and a test, plus a safe order for making changes. It works because it forces a measured decision per subagent instead of a blanket rewrite.

Found on GitHub, shared by anthropics