estimate— agent skill

When this skill is invoked:

1. Read the task description from the argument. If the description is too

vague to estimate meaningfully, ask for clarification before proceeding.

1. Read CLAUDE.md for project context: tech stack, coding standards,

architectural patterns, and any estimation guidelines.

1. Read relevant design documents from design/docs/ if the task relates

to a documented feature or system.

1. Scan the codebase to understand the systems affected by this task:

• Identify files and modules that would need to change
• Assess the complexity of those files (size, dependency count, cyclomatic

complexity)

• Identify integration points with other systems
• Check for existing test coverage in the affected areas

1. Read past sprint data from production/sprints/ if available:

• Look for similar completed tasks and their actual effort
• Calculate historical velocity (planned vs actual)
• Identify any estimation bias patterns (consistently over or under)

1. Analyze the following factors:

Code Complexity:

• Lines of code in affected files
• Number of dependencies and coupling level
• Whether this touches core/engine code vs leaf/feature code
• Whether existing patterns can be followed or new patterns are needed

Scope:

• Number of systems touched
• New code vs modification of existing code
• Amount of new test coverage required
• Data migration or configuration changes needed

Risk:

• New technology or unfamiliar libraries
• Unclear or ambiguous requirements
• Dependencies on unfinished work
• Cross-system integration complexity
• Performance sensitivity

1. Generate the estimate:

## Task Estimate: [Task Name]
Generated: [Date]

### Task Description
[Restate the task clearly in 1-2 sentences]

### Complexity Assessment

| Factor | Assessment | Notes |
|--------|-----------|-------|
| Systems affected | [List] | [Core, business, UI, etc.] |
| Files likely modified | [Count] | [Key files listed below] |
| New code vs modification | [Ratio, e.g., 70% new / 30% modification] | |
| Integration points | [Count] | [Which systems interact] |
| Test coverage needed | [Low / Medium / High] | [Unit, integration, manual] |
| Existing patterns available | [Yes / Partial / No] | [Can follow existing code or new ground] |

**Key files likely affected:**
- `[path/to/file1]` -- [what changes here]
- `[path/to/file2]` -- [what changes here]
- `[path/to/file3]` -- [what changes here]

### Effort Estimate

| Scenario | Days | Assumption |
|----------|------|------------|
| Optimistic | [X] | Everything goes right, no surprises, requirements are clear |
| Expected | [Y] | Normal pace, minor issues, one round of review feedback |
| Pessimistic | [Z] | Significant unknowns surface, blocked for a day, requirements change |

**Recommended budget: [Y days]**

[If historical data is available: "Based on [N] similar tasks that averaged
[X] days actual vs [Y] days estimated, a [correction factor] adjustment has
been applied."]

### Confidence: [High / Medium / Low]

**High** -- Clear requirements, familiar systems, follows existing patterns,
similar tasks completed before.

**Medium** -- Some unknowns, touches moderately complex systems, partial
precedent from previous work.

**Low** -- Significant unknowns, new technology, unclear requirements, or
cross-cutting concerns across many systems.

[Explain which factors drive the confidence level for this specific task.]

### Risk Factors

| Risk | Likelihood | Impact | Mitigation |
|------|-----------|--------|------------|
| [Specific risk] | [High/Med/Low] | [Days added if realized] | [How to reduce] |
| [Another risk] | [Likelihood] | [Impact] | [Mitigation] |

### Dependencies

| Dependency | Status | Impact if Delayed |
|-----------|--------|-------------------|
| [What must be done first] | [Done / In Progress / Not Started] | [How it affects this task] |

### Suggested Breakdown

| # | Sub-task | Estimate | Notes |
|---|----------|----------|-------|
| 1 | [Research / spike] | [X days] | [If unknowns need investigation first] |
| 2 | [Core implementation] | [X days] | [The main work] |
| 3 | [Integration with system X] | [X days] | [Connecting to existing code] |
| 4 | [Testing and validation] | [X days] | [Writing tests, manual verification] |
| 5 | [Code review and iteration] | [X days] | [Review feedback, fixes] |
| | **Total** | **[Y days]** | |

### Historical Comparison
[If similar tasks exist in sprint history:]

| Similar Task | Estimated | Actual | Relevant Difference |
|-------------|-----------|--------|-------------------|
| [Past task 1] | [X days] | [Y days] | [What makes it similar/different] |
| [Past task 2] | [X days] | [Y days] | [What makes it similar/different] |

### Notes and Assumptions
- [Key assumption that affects the estimate]
- [Another assumption]
- [Any caveats about scope boundaries -- what is included vs excluded]
- [Recommendations: e.g., "Consider a spike first if requirement X is unclear"]

1. Output the estimate to the user with a brief summary: recommended

budget, confidence level, and the single biggest risk factor.

Guidelines

• Always give a range (optimistic / expected / pessimistic), never a single

number. Single-point estimates create false precision.

• The recommended budget should be the expected estimate, not the optimistic

one. Padding is not dishonest -- it is realistic.

• If confidence is Low, recommend a time-boxed spike or prototype before

committing to the full estimate.

• Be explicit about what is included and excluded. Scope ambiguity is the

most common source of estimation error.

• Round to half-day increments. Estimating in hours implies false precision

for tasks longer than a day.

• If the task is too large to estimate confidently (more than 10 days

expected), recommend breaking it into smaller tasks and estimating those individually.

• Do not pad estimates silently. If risk exists, call it out explicitly in

the risk factors section so the team can decide how to handle it.

Protocol

• Question: Reads task description from argument; asks for clarification if scope is too vague to estimate
• Options: Skip — single analysis path
• Decision: Skip — estimate is advisory
• Draft: Full estimate shown in conversation only
• Approval: Skip — read-only; no files written

Output

Deliver exactly:

• Estimate: X days (rounded to half-day increments)
• Confidence: High / Medium / Low with one-sentence reason
• Risk factors: explicit list — no silent padding
• Recommendation: proceed / run a spike first / break into smaller tasks