cost-tracking
Per-task cost accounting protocol. Logs token usage and USD cost to the SQLite database for every task, enabling budget control, model comparison, and cost efficiency analysis by the Harness Critic and Evolution Manager.
- License 许可
- MIT
- 最近更新
- 2026-10-18
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- cost-tracking
- agents
- [ "orchestrator", "evolution-manager", "harness-critic" ]
- trigger
- After every task completes (any verdict). Mandatory in all modes.
- description
- Per-task cost accounting protocol. Logs token usage and USD cost to the SQLite database for every task, enabling budget control, model comparison, and cost efficiency analysis by the Harness Critic and Evolution Manager.
03 SKILL.md 正文
Cost Tracking Skill
Why Track Cost
Without cost data:
- Cannot enforce budget_tokens stop rules
- Cannot compare model efficiency (o3 vs Opus vs Sonnet)
- Cannot calculate ROI per finding
- Cannot optimize scanner mode model assignments
- Harness Critic has no efficiency data
What to Track Per Task
After each task completes (any verdict: CONFIRMED, REFUTED, INCONCLUSIVE, KILLED):
task_id: string
tokens_input: int (tokens in the task prompt + context)
tokens_output: int (tokens in the reasoning output)
tokens_total: int (input + output)
cost_usd: float (calculated from model-config.yml pricing)
model_used: string (actual model name, not alias)
phase: string (from task.phase)
verdict: string
over_budget: bool (tokens_total > task.budget_tokens)
Cost Calculation
From model-config.yml, read the active provider's pricing:
# model-config.yml pricing section (per 1M tokens)
providers:
anthropic:
pricing:
claude-opus-4:
input_per_1m: 15.00
output_per_1m: 75.00
claude-sonnet-4:
input_per_1m: 3.00
output_per_1m: 15.00
claude-haiku-3:
input_per_1m: 0.25
output_per_1m: 1.25
openai:
pricing:
o3:
input_per_1m: 10.00
output_per_1m: 40.00
gpt-4.1:
input_per_1m: 2.00
output_per_1m: 8.00
# Calculate cost after each task
cost_usd=$(echo "scale=6; ($input_tokens * $input_price + $output_tokens * $output_price) / 1000000" | bc)
DB Write Protocol
After each task, the Orchestrator writes to the DB:
UPDATE tasks
SET tokens_used = {total_tokens},
cost_usd = {cost_usd},
status = '{verdict}',
completed_at = '{timestamp}',
model_reasoner = '{model_name}'
WHERE task_id = '{task_id}';
-- Also update the run total
UPDATE runs
SET total_cost_usd = total_cost_usd + {cost_usd},
total_tokens = total_tokens + {total_tokens}
WHERE run_id = '{run_id}';
Budget Enforcement
The Orchestrator enforces the budget_tokens field from each task:
Before routing to Deep Reasoner:
- Set context limit = task.budget_tokens
- Include in Reasoner prompt: "STOP if you have used {budget_tokens} tokens.
Apply your stop_rule and output current status."
After Reasoner completes:
- If tokens_used > budget_tokens × 1.2 (20% overrun): flag as over_budget
- Log to Harness Critic: task over-budget by {N} tokens
- This is a harness discipline failure, not a finding failure
Cost Efficiency Metrics (for Harness Critic)
-- Cost per confirmed finding
SELECT
run_id,
SUM(cost_usd) as total_cost,
COUNT(CASE WHEN verdict = 'CONFIRMED' THEN 1 END) as confirmed,
ROUND(SUM(cost_usd) / NULLIF(COUNT(CASE WHEN verdict = 'CONFIRMED' THEN 1 END), 0), 2)
as cost_per_confirmed_finding
FROM tasks
GROUP BY run_id;
-- Most expensive tasks (potential over-analysis)
SELECT task_id, target, cost_usd, tokens_used, budget_tokens, verdict
FROM tasks
ORDER BY cost_usd DESC
LIMIT 10;
-- Model cost comparison across runs
SELECT model_reasoner, AVG(cost_usd) as avg_cost, COUNT(*) as task_count,
COUNT(CASE WHEN verdict = 'CONFIRMED' THEN 1 END) as confirmed
FROM tasks
GROUP BY model_reasoner;
-- Over-budget rate per mode
SELECT r.mode, COUNT(*) as tasks,
SUM(CASE WHEN t.tokens_used > t.budget_tokens THEN 1 ELSE 0 END) as over_budget,
ROUND(100.0 * SUM(CASE WHEN t.tokens_used > t.budget_tokens THEN 1 ELSE 0 END) / COUNT(*), 1)
as over_budget_pct
FROM tasks t JOIN runs r ON t.run_id = r.run_id
GROUP BY r.mode;
Dashboard Cost Section
The Orchestrator updates findings/DASHBOARD.md cost section after each task:
## Budget Status
- Total cost so far: ${total_cost_usd}
- Budget remaining: ${budget_remaining} ({pct}%)
- Cost per resolved task: ${avg_cost_per_task}
- Cost per confirmed finding: ${cost_per_finding}
- Over-budget tasks: {N} ({pct}%)
04 同仓库更多 Skills
stripe-patterns
暂无简介
payment-patterns
暂无简介
testing-patterns
Testing patterns for Jest and Playwright. Use when writing tests, setting up test fixtures, or validating RLS enforcement. Routes to existing test conventions.
team-coordination
Agent Teams orchestration patterns for multi-agent SAFe workflows. Use when spawning agent teams, coordinating teammates, enforcing quality gates via task dependencies, or orchestrating the 11-agent SAFe pipeline. Covers team creation, messaging, shared task lists, and SAFe gate enforcement. Do NOT use for single-agent or simple subagent workflows.
task-compilation
Compile broad audit goals into structured proof obligations with 9 mandatory fields. Used after decomposition by the Threat Modeler agent.
storage-layout-analysis
Proxy patterns, slot collisions, upgrade safety. Used during decomposition to identify storage-related vulnerabilities in upgradeable contracts.