math-progress-monitor
npx machina-cli add skill parcadei/Continuous-Claude-v3/math-progress-monitor --openclawMath Progress Monitor
When to Use
Trigger on phrases like:
- "am I on the right track"
- "is this approach working"
- "I'm stuck"
- "should I try something else"
- "verify my progress"
- "check my reasoning"
- "is this getting too complicated"
Use mid-work to assess whether to continue, pivot, or decompose (Schoenfeld's metacognitive control).
Process
Run a structured progress assessment:
1. Inventory attempts
Ask: "What have you tried so far?"
- List each approach
- Order by when attempted
- Note time spent
2. Extract learnings
Ask: "What did each attempt tell you?"
- Even failures provide information
- What was ruled out?
- What patterns emerged?
3. Complexity check
Ask: "Is complexity growing faster than expected?"
- Warning signs:
- More terms than you started with
- New variables appearing
- Calculation getting messier
- Normal: complexity stays flat or decreases
4. Spot-check verification
Ask: "Can you verify any intermediate results?"
- Run numerical examples
- Check limiting cases
- Dimensional analysis
5. Decomposition check
Ask: "Is there a simpler sub-problem to solve first?"
- Identify dependencies
- Find stepping stones
- Reduce scope
6. Sunk cost assessment
Ask: "What would abandoning this approach cost?"
- Time already spent (irrelevant)
- Insights that transfer (relevant)
- Probability of success if continue
Key Questions to Ask
- What have you tried? (inventory)
- What did you learn? (extract)
- Is it getting simpler or harder? (complexity)
- Can you check a piece? (verify)
- What's the smallest step? (decompose)
- Should you pivot? (decide)
Output Format
Progress Assessment:
Status: [On track / Hitting resistance / Should pivot]
Attempts Summary:
1. [Approach] -> [Outcome/Learning]
2. [Approach] -> [Outcome/Learning]
Complexity Check:
- Expected: [what should this look like]
- Actual: [what it actually looks like]
- Verdict: [Normal / Warning / Abandon]
Verification Points:
- [Intermediate result to check]
- [Numerical test case]
Recommendation:
[Continue / Modify approach / Switch strategy / Decompose / Get more info]
If Pivoting:
- Alternative: [specific new approach]
- Why it might work: [reasoning]
Example Interaction
User: I've been trying to evaluate this integral for an hour
Claude: Let's do a progress check.
Q1: What have you tried?
> u-substitution with u = sin(x), then integration by parts twice.
Q2: What did each tell you?
> Substitution made it longer. Parts created a cycle.
Q3: Complexity now vs start?
> Much worse. Started with 3 terms, now have 8.
Complexity Check:
- Expected: Should simplify or stay similar
- Actual: 3 terms -> 8 terms after 2 techniques
- Verdict: WARNING - likely wrong approach
Let me verify an intermediate step...
[Runs sympy on original integral]
This integral has closed form using:
- Beta function (special function)
- Or contour integration
Recommendation: PIVOT
Your algebraic approaches won't simplify this pattern.
Alternative Approach:
1. Recognize: integral of sin^m(x) * cos^n(x)
2. Use Beta function identity
3. Or convert to complex exponentials
Related Skills:
- complex-analysis/contour-integrals
- numerical-methods/numerical-integration
Would you like the Beta function approach, or get a numerical answer first to verify?
Co-activates
When user is stuck, suggest:
- math-intuition-builder - reset understanding
- math-model-selector - if framework was wrong
- math-mode - for verification computations
- Relevant domain skill - for alternative approach
Source
git clone https://github.com/parcadei/Continuous-Claude-v3/blob/main/.claude/skills/math/math-progress-monitor/SKILL.mdView on GitHub Overview
Math Progress Monitor provides structured mid-work meta-cognition to decide whether to persist, pivot, or decompose. It follows Schoenfeld’s metacognitive control by guiding you through a six-step progress assessment and clear decision logic.
How This Skill Works
During problem solving, trigger phrases prompt a structured progress assessment. It runs through inventorying attempts, extracting learnings, checking complexity, verifying results, decomposing if needed, and assessing sunk costs, then outputs a recommended action and alternative strategies.
When to Use It
- I’m not sure if I’m on the right track or if the approach is valid
- The solution feels longer or more complex than expected
- I’m stuck and need to decide between persisting or pivoting
- I want to verify intermediate results with tests or limits
- I should decompose the problem into smaller sub-problems
- I want to assess time spent vs. potential gains
Quick Start
- Step 1: Trigger a progress check when sensing doubt or complexity rising
- Step 2: Inventory attempts and time spent, then extract what each try reveals
- Step 3: Decide on pivot, decomposition, or continuation, and implement the chosen path
Best Practices
- Ask: What have you tried so far? Inventory each approach with approximate time spent
- Extract learnings from each attempt, including what was ruled out and patterns observed
- Run a complexity check to detect growth in terms/variables or messy calculations
- Verify intermediate results using numerical tests, limiting cases, or dimensional checks
- Consider a simpler sub-problem first and reassess before continuing
Example Use Cases
- Evaluating a difficult integral and pivoting to a Beta function or numerical verification after term growth
- Solving a multi-variable optimization where additional variables emerge making the path impractical
- A geometry problem where decomposing into smaller triangles simplifies the reasoning
- A differential equation approach where verifying a limiting case reveals a misstep
- A combinatorics problem where checking patterns and counterexamples guides the strategy