Task Estimator Skill

Estimates the compute tokens an AI agent needs to complete a TaskMarket task, then calculates cost vs reward to determine if the bounty is profitable.

Mental Model

A task completion has four compute phases:

Phase	What happens	Token weight
Read	Read task description, spec files, existing code	Low
Plan	Reason about approach, architecture decisions	Medium
Execute	Write code/content, make tool calls, iterate	High
Verify	Tests, deployment, submission formatting	Medium

Each phase has input tokens (context) and output tokens (generated text/code). Use the reference tables in references/token-tables.md to score each phase.

Dynamic Context and Arguments

Task text / scope: $ARGUMENTS Task ID (positional): $0

When a task ID is provided, inject live context before scoring:

Task snapshot: !taskmarket task get $0 Board snapshot (batch mode): !taskmarket task list --status open

Estimation Process

1. Classify the task type from $ARGUMENTS or live task context from taskmarket task get $0
2. Score each phase using references/token-tables.md with complexity inferred from $ARGUMENTS/$0
3. Sum tokens → apply model cost → compare to reward
4. Output the estimate in the standard format, including assumptions derived from $ARGUMENTS

Task Type Reference

Type	Examples	Rough total tokens
api-agent	Build a paid Lucid Agent with x402, deploy to Railway	180k–350k
pr-code	Add a feature/fix to an existing repo, open PR	120k–250k
pr-tests	Write test suite for existing package	80k–180k
pr-docs	Write tutorial, README, migration guide	60k–120k
pr-refactor	Port/migrate code, update imports, cleanup	40k–100k
deploy-config	Dockerfile, CI yml, Railway/Vercel config	30k–80k
content	Blog post, SEO article, writeup	40k–90k
research	Data analysis, competitive research	50k–120k
blueprint	Design doc, spec, architecture plan	30k–70k

Standard Output Format

Always produce this block:

## Task Estimate: <task title or first 60 chars>

**Type:** <task-type>
**Complexity:** <low|medium|high|very-high>

### Token Budget
| Phase    | Input tokens | Output tokens | Total   |
|----------|-------------|---------------|---------|
| Read     | X,000       | —             | X,000   |
| Plan     | X,000       | X,000         | X,000   |
| Execute  | X,000       | X,000         | X,000   |
| Verify   | X,000       | X,000         | X,000   |
| **Total**| **X,000**   | **X,000**     | **X,000** |

### Cost Estimate (sonnet-4.5 @ $3/$15 per M)
- Input cost:  $X.XX
- Output cost: $X.XX
- **Total compute cost: $X.XX**

### Reward Analysis
- Task reward: $X.XX USDC
- Compute cost: $X.XX
- Net margin: $X.XX (XX%)
- **Verdict:** <PROFITABLE / BREAK-EVEN / LOSS / SKIP>

### Notes
- <key assumptions or risk factors>
- <what would increase token usage>

Verdict Thresholds

Margin	Verdict
> 70%	PROFITABLE — strong take
40–70%	PROFITABLE — reasonable
10–40%	BREAK-EVEN — marginal, proceed if strategic
< 10%	LOSS — skip unless relationship/reputation value

Model Cost Reference (per 1M tokens)

Model	Input	Output	Notes
claude-sonnet-4.5	$3.00	$15.00	Default workhorse
claude-opus-4	$15.00	$75.00	Complex reasoning
gpt-5-codex	$3.00	$15.00	Code tasks
gemini-2.5-pro	$1.25	$10.00	Budget option

Default to sonnet-4.5 unless task requires deep reasoning (use opus) or pure coding (use codex).

Prices are estimates and can change. Verify live rates for claude-sonnet-4.5, claude-opus-4, gpt-5-codex, and gemini-2.5-pro before final decisions. gpt-5-codex pricing is not consistently published and must be confirmed directly with the provider before use.

Live-rate lookup example: !fetch-model-pricing claude-sonnet-4.5 claude-opus-4 gpt-5-codex gemini-2.5-pro

Key Multipliers

Apply these on top of base estimates:

• TDD required (tests-first): ×1.4 on Execute phase
• Deploy to Railway: +15k tokens (fixed)
• xgate.run listing: +8k tokens (fixed)
• PR + CI green required: +20k tokens (fixed)
• No existing codebase to reference: ×1.2 on Read phase
• Complex external API integration: ×1.3 on Execute
• Multiple deliverables (e.g., API + frontend + tests): ×1.5 overall
• Iteration required (likely back-and-forth): ×1.3 overall

Batch Estimation

When estimating multiple tasks (e.g., reviewing a TaskMarket board), produce a summary table first, then detailed breakdowns on request:

| Task | Type | Tokens | Cost | Reward | Margin | Verdict |
|------|------|--------|------|--------|--------|---------|
| ... |

Sort by margin descending so best opportunities surface first.

Detailed Token Tables

For per-phase token scoring details: see references/token-tables.md For worked examples (api-agent, pr-code, content): see references/examples.md