KeiSeiKit-1.0/skills/session-budget/SKILL.md
Parfii-bot 0be354a920 KeiSeiKit-public — clean state
Single-commit clean baseline after security scrub of niche-tells,
project codenames, internal jargon, and contributor-email leaks.

Contents:
- 100 Rust crates (_primitives/_rust/)
- 37 agent manifests (_manifests/) + generated specs (_generated/)
- 67 user-invocable skills (skills/)
- 33 hooks (hooks/)
- Composition blocks (_blocks/)
- Documentation (docs/, README.md)
- TS adapter packages (_ts_packages/)
- Assembler (_assembler/)
- Roles (_roles/)
- Templates (_templates/)
- Forgejo CI (.forgejo/)

Author: Denis Parfionovich <info@greendragon.info>

License: see LICENSE.
2026-05-01 12:09:03 +08:00

55 lines
1.5 KiB
Markdown

---
name: session-budget
description: Use when tracking token/cost budget per session — monitors API usage and compute costs
arguments:
- name: action
description: "Action: start, check, report, set-limit"
required: true
- name: limit
description: "Budget limit in USD (for set-limit action)"
required: false
---
# Session Budget Tracker
## Actions
### start
Initialize budget tracking for this session:
- Note session start time
- Record any known API costs (Modal, fal.ai, Apify, etc.)
- Set default limit: $5 unless overridden
### check
Before any paid API call:
- Calculate estimated cost of the operation
- Compare against remaining budget
- If over budget: WARN user and ask for confirmation
- If under budget: proceed and log the cost
### report
Generate session cost report:
- List all API calls made and their costs
- Total spend this session
- Remaining budget
- Comparison to previous sessions (if available in memory)
### set-limit
Set custom budget limit for the session:
- Store limit value
- Warn at 80% usage
- Block at 100% unless user overrides
## Cost Reference (approximate)
| Service | Unit | Cost |
|---------|------|------|
| Modal GPU (A10G) | per hour | $1.10 |
| Modal GPU (A100) | per hour | $3.73 |
| fal.ai Flux Pro | per image | $0.05 |
| fal.ai Kling | per video | $0.30-0.90 |
| Apify Actor | per run | varies |
| ElevenLabs TTS | per 1K chars | $0.30 |
## Integration
- Works with api-cost-guard rule (validates before compute)
- Saves session summary to context-store at session end