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;
preferences_root is not a validated pack interface or a required default.
Read testing decision methods 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.
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
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:
- Read relevant TA contracts/NFRs, DA query/migration invariants, SC threats and DH
SLO/recovery criteria. Reuse only evidence for unchanged relevant inputs.
- 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.
- 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.
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.
1---2name: li-tq3description: 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.4metadata:5 source: https://skillmd.com/skills/jokerman89/li-tq6---78> **Lintel on GitHub Copilot.** Generated from `skills/tq/SKILL.md`; edit the canonical file, then run9> `li-copilot init`.10> - **Resource root:** `../../..` from this skill's base directory (the Lintel source with `bin/`,11> `lib/`, `skills/`). Write plans, state and evidence into the working repository's `.claude/`12> tree, never into the resource root.13> - **Skill-relative paths:** this skill's own `scripts/`, `references/` and `data/` folders (and a14> `<base>` that the workflow defines as its own directory) mean15> `../../../skills/tq/` in the Lintel source, not this generated folder.16> `${LINTEL_SKILLS_DIR:-skills}` means the skills root, `../../../skills`. A `bin/li-run` step17> runs in the working repository, so use `$LINTEL_SKILLS_DIR/tq/` there.18> - **Shell steps:** run Bash snippets with Bash (Git for Windows' `bash.exe` on Windows, never19> `System32\bash.exe`). Save a snippet to a temporary `.sh` file and run20> `bash "<resource root>/bin/li-run" <file>`; it prepares `LINTEL_SOURCE_ROOT`, `LINTEL_REPO_ROOT`21> and the profile context.22> - **Tools:** Read=`view`, Write=`create`, Edit=`edit`, Bash=`bash`/`powershell`, Grep=`grep`,23> Glob=`glob`, AskUserQuestion=`ask_user`, TodoWrite=the plan checklist, Task or a named role=`task`24> with that custom agent, WebFetch=`web_fetch`.25> - **Other Lintel workflows** are native skills: invoke `/li-<name>` rather than reading their26> files. Named roles such as `CodeReviewer` are custom agents.2728# Testing and quality assurance2930Read the [optional preference metadata clarification](../../../skills/da/references/preferences.md);31`preferences_root` is not a validated pack interface or a required default.3233Read [testing decision methods](../../../skills/tq/references/decision-methods.md) for consumer-specific34compatibility, comparable baselines, uncertainty and actual runtime evidence.35`/li-tq full` covers all checkpoints; saved `/li-tq loop` revisits the affected strategy/results;36`/li-tq <capability>` or `/li-tq single --action <capability>` stays targeted. Ordinary test execution can37use the existing runner/QA workflow without creating a domain plan.3839## Sub-capability dispatch4041| Capability | Dispatches to (agents) | Produces | Decisions and checks |42|---|---|---|---|43| `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 |44| `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 |45| `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 |46| `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 |47| `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 |48| `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 |49| `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 |5051Existing tooling can include `go test -coverprofile`, `pytest --cov`, the project's52Jest/Vitest coverage, `cargo tarpaulin` or equivalent. Inspect manifests and effects53first; do not install a guessed tool, use production services or claim unrun checks.5455Performance consumers use the one56[comparable performance-evidence method](../../../skills/tq/references/decision-methods.md#comparable-performance-evidence).57Choose one analyzer, keep actual samples/sampling/tail uncertainty and pass its exact58source/workload/window artifact to the read-only budget view. Do not rerun a profiler59to reformat that result. CapacityPlanner and SystemArchitect retain projection and60invariant/NFR purposes, not another default collection pass.6162## Workflow6364Read and execute the [shared module caller procedure](../../../skills/full-engineering-pass/references/domain-handoff.md#module-caller-procedure)65before domain work. It is the sole owner of original work/package/leaf admission,66live P07/policy checks, immutable obligations, checkpoint publication, cold continuation67and final QA/independent acceptance. Apply this testing method inside that procedure:68691. Read relevant TA contracts/NFRs, DA query/migration invariants, SC threats and DH70 SLO/recovery criteria. Reuse only evidence for unchanged relevant inputs.712. Use the capability and checkpoint tables to choose requirement assertions, actual72 consumer/provider cases and comparable measurements. TestRunner runs/classifies/73 reports without fixing; it is independent only when actually separate.743. Execute inspected, authorized checks and retain stdout/stderr, counts, skips,75 exits, source/environment and limitations. ContractTestArchitect designs; real76 runners verify. A toy example or grep is not live compatibility evidence.7778## Checkpoint ownership7980| Checkpoint | Method/owner | Observable acceptance |81|---|---|---|82| `coverage_targets_met` | TestRunner measures, Architect maps risk | exact requirement-to-assertion gaps and applicable targets, not coverage alone |83| `perf_budgets_locked` | PerfBudgetEnforcer consumes the selected analyzer's evidence | baseline/candidate identity, sample distributions, meaningful thresholds and real positive/negative gate evidence |84| `contract_tests_complete` | APIDesigner + ContractTestArchitect + actual runner | each supported consumer/version checked; required failed/unknown consumers remain blockers |85| `regression_suite_curated` | RegressionDetective + TestRunner | reproduced failure, retained fix-to-test link, golden and edge paths, flake disposition |86| `chaos_scenarios_documented` | SecurityAuditor + SystemArchitect | scenario/control/abort/owner plan; actual recovery if required, otherwise explicitly unexecuted design |8788Worked contrast: adding `paused` to an enum breaks a closed generated consumer but89may work with a tested unknown-value branch. Similarly, a single faster benchmark90run does not prove improvement: retain variance/environment, offered load and tail91sample limits. Production fault injection always needs its own exact authority.9293## Handoff, loop and honest failure9495Preserve coverage, performance, contract/version matrix, regression, chaos,96quarantine and test-pyramid artifacts. Bind actual commands/output to original97requirements, receiver mode and safe operation/iteration paths. Existing98`state/tq/` reports remain available when explicitly selected, not newest-file truth.99100A failed or interrupted runner stays failed/incomplete. Do not drop NaNs, skipped101scenarios or timeouts from the denominator and call the remainder a required pass.102Cold resume uses the shared state table; do not automatically repeat a side-effecting103test or erase an earlier failure. A loop creates new evidence for affected inputs.104105Six advisory dimensions remain Critical-path coverage, Perf budgets, Contract tests,106Regression suite, Chaos scenarios and Test pyramid, measured against the observable107checkpoint criteria. Missing/failed mandatory evidence blocks despite a108high average. DONE requires actual selected checks and independent acceptance;109DONE_WITH_CONCERNS retains only advisory issues; BLOCKED/NEEDS_CONTEXT names required110gaps. Module completion is not SHIP permission or enclosing-task completion.111112## Reusable patterns113114Follow the [reusable pattern consumer contract](../../../skills/pattern/references/consumer-contract.md).115Module entry resolves the testing and quality expectations bound to the current target context116and passes each sub-capability only its projected clauses (`project` for the mapped package, or117the clauses mapped to it). An unknown target or quality tier yields `needs-context` and blocks118the dependent design; no live cloud, tenant or environment discovery is implied, and119requirements are never inferred from generic recommendations. When the runtime reports no120patterns, the module is unchanged.121122## Integration and dormant hooks123124TQ validates the promises of TA/DA/SC/DH; it does not rewrite their schemas or policy.125It is the final engineering-domain module in composition order, not proof that all126client/installed or compound acceptance is complete.127Optional audit/Brief Forge use remains explicit. Preserve opt-in ADR-0008 hooks:128129- `hooks/shared/tq-coverage-drop-warn/`130- `hooks/shared/tq-perf-regression-warn/`131- `hooks/shared/tq-contract-break-warn/`132133These optional warnings are unobserved without actual execution evidence. The hook134file is not proof of registration, enforcement or a successful test run, and cannot135replace an applicable required check.