# Li Tq

> Use for testing and QA-strategy depth — test-pyramid review, coverage audits, contract-test design, regression suites, flaky-test quarantine, perf budgets, and chaos plans. Reach for it when test strategy needs deliberate design rather than ad-hoc tests. Runs full, loop, or single-capability, dispatching to the test agents and scoring against a rubric.

- Skill: `jokerman89/li-tq` (Agent Skill)
- Install (CLI): `npx get-skill add jokerman89/li-tq`
- Raw SKILL.md: https://api.skillmd.com/api/skills/jokerman89/li-tq/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: jokerman89 (https://skillmd.com/u/jokerman89)
- Updated: 2026-11-15
- Page: https://skillmd.com/skills/jokerman89/li-tq

---


> **Lintel on GitHub Copilot.** Generated from `skills/tq/SKILL.md`; edit the canonical file, then run
> `li-copilot init`.
> - **Resource root:** `../../..` from this skill's base directory (the Lintel source with `bin/`,
>   `lib/`, `skills/`). Write plans, state and evidence into the working repository's `.claude/`
>   tree, never into the resource root.
> - **Skill-relative paths:** this skill's own `scripts/`, `references/` and `data/` folders (and a
>   `<base>` that the workflow defines as its own directory) mean
>   `../../../skills/tq/` in the Lintel source, not this generated folder.
>   `${LINTEL_SKILLS_DIR:-skills}` means the skills root, `../../../skills`. A `bin/li-run` step
>   runs in the working repository, so use `$LINTEL_SKILLS_DIR/tq/` there.
> - **Shell steps:** run Bash snippets with Bash (Git for Windows' `bash.exe` on Windows, never
>   `System32\bash.exe`). Save a snippet to a temporary `.sh` file and run
>   `bash "<resource root>/bin/li-run" <file>`; it prepares `LINTEL_SOURCE_ROOT`, `LINTEL_REPO_ROOT`
>   and the profile context.
> - **Tools:** Read=`view`, Write=`create`, Edit=`edit`, Bash=`bash`/`powershell`, Grep=`grep`,
>   Glob=`glob`, AskUserQuestion=`ask_user`, TodoWrite=the plan checklist, Task or a named role=`task`
>   with that custom agent, WebFetch=`web_fetch`.
> - **Other Lintel workflows** are native skills: invoke `/li-<name>` rather than reading their
>   files. Named roles such as `CodeReviewer` are custom agents.

# Testing and quality assurance

Read the [optional preference metadata clarification](../../../skills/da/references/preferences.md);
`preferences_root` is not a validated pack interface or a required default.

Read [testing decision methods](../../../skills/tq/references/decision-methods.md) for consumer-specific
compatibility, comparable baselines, uncertainty and actual runtime evidence.
`/li-tq full` covers all checkpoints; saved `/li-tq loop` revisits the affected strategy/results;
`/li-tq <capability>` or `/li-tq single --action <capability>` stays targeted. Ordinary test execution can
use the existing runner/QA workflow without creating a domain plan.

## Sub-capability dispatch

| Capability | Dispatches to (agents) | Produces | Decisions and checks |
|---|---|---|---|
| `coverage-audit` | TestRunner + Architect | per-component requirement/line/branch/mutation gaps and backfill priorities | approved critical-path targets, observed commands/counts and exclusions; no universal 80/100% acceptance |
| `perf-budget-spec` | PerfBudgetEnforcer; one PerformanceAnalyzer or LatencyAnalyzer only for missing evidence | per-journey budgets, detection and CI-gate design | consume the same comparable evidence, then propose budgets; no default parallel profiling; a spec is not enforcement |
| `contract-test-design` | APIDesigner + ContractTestArchitect | interface surface, consumer expectations and version matrix | actual serializers and active consumers; REST/GraphQL/gRPC/events/IPC as applicable; no mandatory Pact |
| `regression-suite` | RegressionDetective + TestRunner | fix-to-test mapping and golden-path execution health | select relevant history explicitly; isolated bisection with original-state protection, not automatic caller changes |
| `chaos-plan` | SecurityAuditor + SystemArchitect | failure matrix, abort/recovery criteria and owners | selected threat/on-call evidence; disabled chaos is a scoped not-run/N/A decision, not a passing stub |
| `flaky-quarantine` | TestRunner + RegressionDetective | repeated-outcome evidence, remediation owner and bounded quarantine | same source/env/seed/order; passing on retry does not erase failure; policy-derived duration and critical coverage |
| `test-pyramid-review` | TestRunner + Architect | test-kind distribution, time, flake rate and rebalancing options | unit/integration/e2e/contract/perf; optimize feedback per risk, not a universal count ratio |

Existing tooling can include `go test -coverprofile`, `pytest --cov`, the project's
Jest/Vitest coverage, `cargo tarpaulin` or equivalent. Inspect manifests and effects
first; do not install a guessed tool, use production services or claim unrun checks.

Performance consumers use the one
[comparable performance-evidence method](../../../skills/tq/references/decision-methods.md#comparable-performance-evidence).
Choose one analyzer, keep actual samples/sampling/tail uncertainty and pass its exact
source/workload/window artifact to the read-only budget view. Do not rerun a profiler
to reformat that result. CapacityPlanner and SystemArchitect retain projection and
invariant/NFR purposes, not another default collection pass.

## Workflow

Read and execute the [shared module caller procedure](../../../skills/full-engineering-pass/references/domain-handoff.md#module-caller-procedure)
before domain work. It is the sole owner of original work/package/leaf admission,
live P07/policy checks, immutable obligations, checkpoint publication, cold continuation
and final QA/independent acceptance. Apply this testing method inside that procedure:

1. Read relevant TA contracts/NFRs, DA query/migration invariants, SC threats and DH
   SLO/recovery criteria. Reuse only evidence for unchanged relevant inputs.
2. Use the capability and checkpoint tables to choose requirement assertions, actual
   consumer/provider cases and comparable measurements. TestRunner runs/classifies/
   reports without fixing; it is independent only when actually separate.
3. Execute inspected, authorized checks and retain stdout/stderr, counts, skips,
   exits, source/environment and limitations. ContractTestArchitect designs; real
   runners verify. A toy example or grep is not live compatibility evidence.

## Checkpoint ownership

| Checkpoint | Method/owner | Observable acceptance |
|---|---|---|
| `coverage_targets_met` | TestRunner measures, Architect maps risk | exact requirement-to-assertion gaps and applicable targets, not coverage alone |
| `perf_budgets_locked` | PerfBudgetEnforcer consumes the selected analyzer's evidence | baseline/candidate identity, sample distributions, meaningful thresholds and real positive/negative gate evidence |
| `contract_tests_complete` | APIDesigner + ContractTestArchitect + actual runner | each supported consumer/version checked; required failed/unknown consumers remain blockers |
| `regression_suite_curated` | RegressionDetective + TestRunner | reproduced failure, retained fix-to-test link, golden and edge paths, flake disposition |
| `chaos_scenarios_documented` | SecurityAuditor + SystemArchitect | scenario/control/abort/owner plan; actual recovery if required, otherwise explicitly unexecuted design |

Worked contrast: adding `paused` to an enum breaks a closed generated consumer but
may work with a tested unknown-value branch. Similarly, a single faster benchmark
run does not prove improvement: retain variance/environment, offered load and tail
sample limits. Production fault injection always needs its own exact authority.

## Handoff, loop and honest failure

Preserve coverage, performance, contract/version matrix, regression, chaos,
quarantine and test-pyramid artifacts. Bind actual commands/output to original
requirements, receiver mode and safe operation/iteration paths. Existing
`state/tq/` reports remain available when explicitly selected, not newest-file truth.

A failed or interrupted runner stays failed/incomplete. Do not drop NaNs, skipped
scenarios or timeouts from the denominator and call the remainder a required pass.
Cold resume uses the shared state table; do not automatically repeat a side-effecting
test or erase an earlier failure. A loop creates new evidence for affected inputs.

Six advisory dimensions remain Critical-path coverage, Perf budgets, Contract tests,
Regression suite, Chaos scenarios and Test pyramid, measured against the observable
checkpoint criteria. Missing/failed mandatory evidence blocks despite a
high average. DONE requires actual selected checks and independent acceptance;
DONE_WITH_CONCERNS retains only advisory issues; BLOCKED/NEEDS_CONTEXT names required
gaps. Module completion is not SHIP permission or enclosing-task completion.

## Reusable patterns

Follow the [reusable pattern consumer contract](../../../skills/pattern/references/consumer-contract.md).
Module entry resolves the testing and quality expectations bound to the current target context
and passes each sub-capability only its projected clauses (`project` for the mapped package, or
the clauses mapped to it). An unknown target or quality tier yields `needs-context` and blocks
the dependent design; no live cloud, tenant or environment discovery is implied, and
requirements are never inferred from generic recommendations. When the runtime reports no
patterns, the module is unchanged.

## Integration and dormant hooks

TQ validates the promises of TA/DA/SC/DH; it does not rewrite their schemas or policy.
It is the final engineering-domain module in composition order, not proof that all
client/installed or compound acceptance is complete.
Optional audit/Brief Forge use remains explicit. Preserve opt-in ADR-0008 hooks:

- `hooks/shared/tq-coverage-drop-warn/`
- `hooks/shared/tq-perf-regression-warn/`
- `hooks/shared/tq-contract-break-warn/`

These optional warnings are unobserved without actual execution evidence. The hook
file is not proof of registration, enforcement or a successful test run, and cannot
replace an applicable required check.

