-
Notifications
You must be signed in to change notification settings - Fork 1
Agent Test Writer
Oak Dev-inter edited this page Apr 26, 2026
·
2 revisions
One target per call. Regression case mandatory after every bug fix.
- User asks for tests on a specific function / class / endpoint
- Immediately after
bug-hunterships a fix (regression test) - Backfilling coverage before
refactor-surgeoncan proceed - New endpoint needs happy-path + edge-case coverage
- Reads target and its neighbors β inputs, outputs, side effects
- Reads existing test files to match framework, style, fixture pattern
- Lists test cases BEFORE writing: happy path / edges (empty, null, zero, max, unicode, TZ, concurrent) / error path / regression
- Writes tests with one assertion focus each, descriptive names
- Runs suite β reports runtime green or why not
- Reports coverage gaps and why deferred
- Mock code under test (mocks only for external HTTP, clock, filesystem)
- Use inline magic fixtures β pulls from project factories / seeders
- Emit snapshot tests for logic (only for stable serialized output)
- Write tests that merely re-type the implementation
- Accept a call that lists multiple targets β scope is one per invocation
- Target:
file:functionor endpoint - Context: what was changed, or what bug was just fixed
- Test framework (optional β detected from project)
TARGET: <file:function>
FRAMEWORK: <pytest | jest | phpunit | ...>
TEST FILE: <path>
CASES WRITTEN:
β happy_path_<desc>
β edge_<desc>
β error_<desc>
β regression_<bug-ref>
RUNTIME: <pass/fail from test runner>
GAPS (intentionally not tested):
- <case> β <why deferred>
HANDOFF: ready for CI / review.
Defaults to Sonnet. Opus rarely needed; tests favor mechanical enumeration.
- β Mocking the function being tested
- β Bug fix without a regression test
- β "assertEquals(impl(), impl())"-style tautologies
- β Snapshot test on business logic output
Introduced in v0.9.1.
- Master-Orchestrator
- Multi-Agent-Orchestration
- Agent-Bug-Hunter
- Agent-Refactor-Surgeon
-
Agent-Audit-Specialist β for
--focus=qualityreview of test code itself (deprecated[[Agent-Code-Reviewer]]stub still works)
Repo β’ Discussions β’ Issues β’ Changelog β’ Security β’ Contributing β’ MIT β’ Β© Kasidit Wansudon
Kasidit
Core
- Commands
- Kasi-Mode π₯ v0.10
- Backend-Hooks π₯ v0.10
- Model Tiers
- Gravity Pattern
- Multi-Agent-Orchestration
- Claude Design Integration
- UI Override Mode
- FAQ
Version History
- Version History β overview
- v0.13.0 β thClaws (Consolidated) π¦
- v0.12.0 β thClaws Runtime Support π¦
- v0.11.0 β Backend + Bridge + Runbook π
- v0.10.0 β Mode + Backend Hooks
- v0.9.2 β Gravity
- v0.9.1 β Master Orchestrator
- v0.9.0 β Claude Design
- v0.8.0 β Tier Cascade
- v0.7.4 β SWE-bench
- v0.3.0 β Tier adaptation
- v0.2.1 β Docs protocol
- v0.2.0 β UI Override
- v0.1.0 β Core
Concepts
Commands
- Kasi-Init
- Kasi-Review
- Kasi-Security
- Kasi-Fix
- Kasi-Ui
- Kasi-Cascade
- Kasi-Multi
- Kasi-Scaffold
- Kasi-Docs
- Kasi-Status
- Kasi-Promote
- Kasi-Pull
- Kasi-Sync
- Kasi-Search
- Kasi-Wiki-Sync
Agents
- Agent-Architect-Planner
- Agent-Audit-Specialist π₯ v0.10
- Agent-Bug-Hunter
- Agent-Deep-Researcher
- Agent-Legacy-Specialist
- Agent-Migration-Specialist
- Agent-Refactor-Surgeon
- Agent-Test-Writer
Deprecated v0.10 (stubs β audit-specialist --focus=..., removed in v0.11)
-
Agent-Code-Reviewer β
--focus=quality -
Agent-Security-Auditor β
--focus=security -
Agent-Perf-Profiler β
--focus=perf