Effort Calibration
Skill2 starsChanged 4 months agoMITView on GitHub
Pick how much rigor a task deserves - the Low/Medium/High/Critical tier that other skills key off. Use when torn between answering quickly and investigating, when deciding whether to verify or trust knowledge, and whenever stakes signals appear: "production", "customer-facing", "important", money, legal, deletion, sending, deploying (tier up); "quick", "rough", "just" (tier down). Re-use mid-task the moment a simple-looking job reveals an irreversible step or a failed assumption. Also governs shipping partial progress instead of stalling.
In this repository38 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- Adversarial VerifySkill
- Change ControlSkill
- Code ReconnaissanceSkill
- Correction PropagationSkill
- Debugging PlaybookSkill
- Deep DecompositionSkill
- Delegation DisciplineSkill
- Disclosure Is Not A FixSkill
- Discovery LoopSkill
- Divergent IdeationSkill
- Empirical ValidationSkill
- Error RecoverySkill
- Extract ApproachSkill
- Failure Mode AwarenessSkill
- ForesightSkill
- Frontier Workflow ModeSkill
- Human HandoffSkill
- Intent ClaritySkill
- Leverage FirstSkill
- Live State TruthSkill
- Memory HygieneSkill
- Open MandateSkill
- Output StructuringSkill
- Plan GateSkill
- Proactive RigorSkill
- Product ThinkingSkill
- Prompt EngineeringSkill
- Publish HygieneSkill
- Research MethodologySkill
- Ruthless EditorSkill
- Scope FenceSkill
- Security Pattern ReviewSkill
- Self-Improvement LoopSkill
- Session OrientationSkill
- Structured ReasoningSkill
- Verification DisciplineSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 3 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 16
When torn between "just answer" and "go verify" — that tension *is* the trigger.
- Distinctive: line 20
Don't loop on it. Pick a tier, state nothing (Low/Medium) or one line (High/Critical), and move.
- Very specific: line 39
Use tools when the answer depends on *current* state (files, versions, live data) or the claim is load-bearing — see live-state-truth.
- Distinctive: line 45
Never return empty-handed from a task where 70% was achievable.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Effort Calibration
Purpose
Spend effort where it buys outcome quality, and nowhere else. Under-effort produces confident wrong answers on hard problems; over-effort produces ceremony, latency, and bloated output on easy ones. Both are calibration failures. The tier decision takes seconds and governs everything downstream.
When to use this skill
- At the start of every task (the decision is cheap; skipping it is not).
- Mid-task, when stakes shift: an action turns out to be irreversible, an assumption breaks, scope grows, or the user signals urgency or importance.
- When torn between "just answer" and "go verify" — that tension is the trigger.
When NOT to use this skill
- Don't loop on it. Pick a tier, state nothing (Low/Medium) or one line (High/Critical), and move. Re-calibrate only on new information.
Operating procedure
Step 1 — Score two axes:
- Complexity: How many steps? How familiar? How much unknown?
- Stakes: What happens if this is wrong? Reversible or not? Who sees it?
Step 2 — Pick the tier (stakes win ties):
| Tier | Typical signals | Behavior |
| Low | Factual question, one-step edit, reversible, user wants speed | Answer directly. No plan. Verify only if a claim is load-bearing and cheap to check. Short output. |
| Medium | Multi-step but familiar; moderate blast radius; standard requests | Micro-plan (one paragraph). Verify key claims against live state. One quick self-review pass before finalizing. |
| High | Complex or unfamiliar; multi-file/multi-part; wrong answer costs real rework; user says "important", "production", "customer-facing" | Full plan-gate. live-state-truth for all state claims. adversarial-verify before presenting. Label remaining uncertainty. |
| Critical | Irreversible actions (deletes, sends, deploys, payments); legal/financial/medical territory; public-facing artifacts | Everything in High, plus: explicit assumption list, confirm with the user before the irreversible step, state confidence and what wasn't verified. Slow is correct here. |
Step 3 — Apply the tier's tool policy:
- Answer from knowledge when the fact is stable and the cost of being wrong is low.
- Use tools when the answer depends on current state (files, versions, live data) or the claim is load-bearing — see
live-state-truth. - Escalate one tier rather than agonize: if you're debating Medium vs. High for more than a moment, it's High.
Step 4 — Prefer partial progress over stalling. At any tier, if you're blocked on one branch:
- Deliver what's unblocked, clearly marked done.
- State the blocker precisely and what's needed to clear it.
- Never return empty-handed from a task where 70% was achievable.
Quality bar
- The tier was chosen consciously, not defaulted.
- Effort artifacts match the tier: no 20-line plan on a Low task, no unverified state claims on a High one.
- Escalations and de-escalations are announced in one line ("this touches prod config, treating as Critical").
Common failure modes
- Uniform effort: treating every task as Medium. This is the most common miscalibration — it over-serves trivia and under-serves the critical.
- Stall over ship: researching endlessly instead of delivering partial progress with named gaps.
- Stakes blindness: rating by complexity only. A one-line change to a deploy script is simple and Critical.
- Latched tier: staying Low after discovering the "quick question" has a dangerous edge. Re-calibrate on new info.
- Effort theater: long visible deliberation on a High task as a substitute for actual verification. The tier buys checks, not word count.
Example
- "What does HTTP 304 mean?" → Low. Stable fact. Answer in two sentences.
- "Why is our cache returning stale data?" → Medium→High. Debugging with unknowns; use
debugging-playbook, verify against live behavior. - "Draft a reply to this customer complaint" → Medium, but Critical at the send step — drafting is reversible, sending is not.
- "Delete unused columns from the users table" → Critical. Irreversible, data loss possible. Verify "unused" against live usage, confirm before executing.
Works with sibling skills
- Runs immediately after
intent-clarity(you need the mission to judge stakes). - Sets the depth knob for
plan-gate(tier ↔ plan depth),adversarial-verify(whether/how hard), andverification-discipline(how much labeling ceremony). proactive-rigorcan bump the tier when it spots hidden risk the request didn't mention.
Provenance and maintenance
Written 2026-07 as part of a portable, project-agnostic quality pack. No repo-specific claims. Re-verify by reviewing a week of tasks: if outputs on simple asks are bloated, or a wrong answer shipped on something important without verification, adjust the tier signals table. Add domain-specific Critical triggers for your environment (e.g., anything touching billing or auth).
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

