qa-context-loader

name	qa-context-loader
description	Detects test framework, loads QA context files, reads agent state and learnings before agent execution
user-invocable	false

QA Context Loader

Before performing your primary task, detect the project's test infrastructure and load QA context.

Steps

1. Detect Test Framework

Check for test framework indicators in order. Stop at the first match.

Node.js / TypeScript:

• Read package.json in project root. Check devDependencies and scripts:
  • vitest in deps or vitest in scripts → Vitest (npx vitest run, coverage: npx vitest run --coverage)
  • jest in deps or jest in scripts → Jest (npx jest, coverage: npx jest --coverage)
  • mocha in deps → Mocha (npx mocha)
• Check for config files: vitest.config.ts, jest.config.js, jest.config.ts, .mocharc.yml

Python:

• Read pyproject.toml or setup.cfg. Check for:
  • pytest in dependencies or [tool.pytest] section → Pytest (pytest, coverage: pytest --cov)
  • unittest patterns → Unittest (python -m unittest discover)
• Check for conftest.py, pytest.ini

Go:

• Check for go.mod → Go test (go test ./..., coverage: go test -coverprofile=coverage.out ./...)

Rust:

• Check for Cargo.toml → Cargo test (cargo test)

Ruby:

• Check for Gemfile. Check for:
  • rspec in Gemfile → RSpec (bundle exec rspec, coverage via simplecov)
  • minitest → Minitest (bundle exec ruby -Itest)

Record the detected framework, test command, and coverage command.

2. Detect Test File Patterns

Based on the framework, identify where tests live:

• Glob for **/*.test.{ts,tsx,js,jsx}, **/*.spec.{ts,tsx,js,jsx} (JS/TS)
• Glob for **/test_*.py, **/*_test.py (Python)
• Glob for **/*_test.go (Go)
• Glob for **/*_spec.rb (Ruby)

Count total test files found.

3. Detect Lint and Type Check Tools

Linting:

• eslint in package.json deps → npx eslint .
• ruff in pyproject.toml → ruff check .
• golangci-lint installed → golangci-lint run
• rubocop in Gemfile → bundle exec rubocop

Type checking:

• typescript in package.json deps → npx tsc --noEmit
• mypy in pyproject.toml → mypy .
• pyright in pyproject.toml → pyright

Record detected tools or "none detected."

4. Load QA Context Files

Look for context/qa/ directory in the project root.

If it exists, read all present files in this order:

File	Purpose	Required By
test-strategy.md	Testing philosophy, what to test vs skip	test-writer
conventions.md	Naming, placement, assertion style, mock patterns, quality bar	test-writer
critical-paths.md	Business-critical code paths with glob patterns	risk-analyzer, coverage-checker
coverage-policy.md	Coverage thresholds by tier, changed-line policy, exemptions	coverage-checker
ownership.md	Who owns test coverage for which code areas, team maturity	test-writer, coverage-checker
frameworks.md	Custom framework config overrides (auto-populated by /scout:init)	all agents
risk-config.yaml	Risk weights per category, large change threshold	risk-analyzer
learnings/what-works.md	QA patterns proven effective	all agents
learnings/what-doesnt.md	QA anti-patterns proven to fail	all agents

For each file:

• Ready — >50 words, <3 placeholder patterns
• Needs work — exists but mostly placeholders
• Missing — file not found

Check for placeholder patterns ([placeholder], [e.g.,, [Add ). Note files that need filling.

If it doesn't exist, proceed with auto-detected defaults. Note that /scout:init can scaffold context files.

5. Load Learnings

If context/qa/learnings/what-works.md exists, read it. Extract patterns relevant to the current agent's task:

• test-runner: Risk detection patterns, flaky test management
• coverage-checker: Coverage strategy learnings
• test-writer: Test writing patterns, mocking strategies
• pre-push-gate: Process learnings
• risk-analyzer: Risk detection patterns

If context/qa/learnings/what-doesnt.md exists, read it. Note anti-patterns to avoid.

If neither exists, proceed without learnings. Note that filling these in improves agent quality over time.

6. Load Agent State

Read context/qa/agent-state.json if it exists. The state file tracks:

Per agent:

• last_run — ISO 8601 timestamp
• run_count — total executions
• last_summary — one-line result from last run

Signals — detected issues that persist across sessions:

• id — unique identifier (format: {type}-{identifier}-{date})
• type — signal category (see below)
• severity — Critical / High / Medium / Low
• description — human-readable explanation
• detected_at — when first detected
• status — active / resolved / dismissed

Signal types for QA agents:

• coverage-declining — coverage trend is going down
• untested-critical-path — critical path code has no tests
• persistent-failure — same test failing across multiple runs
• flaky-test — test that passes/fails non-deterministically
• risk-score-spike — risk score jumped significantly
• new-untested-code — new source files without corresponding tests

Actions taken — log of all executed actions:

• timestamp, action (e.g., run-tests, write-test-file, run-coverage), description

Feedback — user ratings on agent usefulness:

• timestamp, rating (Yes / Partially / No), feedback_note (optional)

For the current agent:

• Extract last_run, run_count, last_summary
• Identify active signals (don't re-alert on same signal)
• Identify dismissed signals (skip entirely)
• Review feedback history (note patterns — if a signal type is consistently rated "No", deprioritize it)
• Review flaky_tests list

If the file does not exist, note this is a first run.

7. Load Autonomy Config

Read context/qa/autonomy.yaml if it exists. For the current agent, summarize which action types are autonomous, requires_approval, or disabled. Default all actions to autonomous (QA actions are safe — they read/run, not deploy).

8. Summarize

Before proceeding with the primary task, briefly report:

• Detected framework and test command
• Number of test files found
• Lint/typecheck tools detected
• Context file status (present/missing/needs-work per file)
• Learnings loaded (count of patterns from what-works + what-doesnt)
• Agent state (last run, active signals, known flaky tests)
• Autonomy config (action permissions for current agent)