--- name: griddynamics/testgen-flow source: https://app.decimal.ai/s/griddynamics-testgen-flow@2/SKILL.md source_sha256: 8c2b1e0a18fc --- Systematic requirements analysis from Issue Tracker tickets and Wiki documentation to structured requirements and test scenarios. Extracts data, identifies gaps, clarifies unknowns via HITL, generates requirements document, and produces test cases with export to a Test Management System (Phase 6, user-triggered — the user may choose not to trigger it, but it is a fully-specified phase, not informally skippable). Designed for BA/QA engineers and requirements engineers. Prerequisite: Rosetta Prep Steps. **Terminology.** External systems are named by role throughout this workflow and its phases: **Issue Tracker**, **Wiki**, and **Test Management System (TMS)**. Jira, Confluence, and TestRail are canonical examples only — adapt identifiers, URLs, requests, calls, and query syntax to the systems resolved for the current project (from repository-root `gain.json`, explicit user input, recognizable URLs/handles, and available integrations). - All Rosetta prep steps MUST be FULLY completed - USE SKILL `load-project-context`, `orchestration`, `hitl` - **ONE PHASE AT A TIME:** APPLY PHASE the phase file, execute, update state, move to next. **JIT**: phase-by-phase: do not load/act IN ADVANCE. - **DO NOT SKIP PHASES:** Each builds on the previous. Skip gates: only with **explicit user instruction**, **or** when `testgen-state.md` marks the phase complete **and** its expected output file exists under `plans/testgen-{TICKET-KEY}/`; otherwise resume from the earliest incomplete phase. The **explicit user instruction** skip NEVER applies to the Phase 3 / Phase 6 HITL gates — those are rule 2 of `` and are never overridden. - **Phase-file load failure:** if APPLY PHASE for a phase file returns nothing, retry once, then HALT and report — do not improvise the phase. - **Transition precedence:** `` priority hierarchy. - **STATE TRACKING:** Update `plans/testgen-{TICKET-KEY}/testgen-state.md` after each phase. - **SELF-CHECK BETWEEN PHASES:** Before advancing, verify the state row was updated, the expected output file exists and is non-empty, the phase's `## Metrics` count is populated (a thin `0`/`1` → re-check the artifact), and any HITL approval (Phase 3, 6) is recorded. - When a phase delegates work to subagents, dispatch per USE SKILL `orchestration`. - MUST ALWAYS use todo tasks ledger, ASAP. Prioritize ACCURACY over SPEED. Proactively use available tools/MCPs, if question can be answered from code or MCP call - do that first. - MUST create output directory `plans/testgen-{TICKET-KEY}/` at start. - **Trigger prompt example:** `Analyze requirements for PROJ-123` (also: bare key `PROJ-123`, full ticket URL). Ticket-only and ticket+Wiki input formats are enumerated in APPLY SKILL FILE `phases/testgen-flow-project-config-loading.md` step 0.1. - **Per-phase failure cases — owned by phase files:** - *Ticket not found* → APPLY SKILL FILE `phases/testgen-flow-data-collection.md` + the `data-collection` skill's Issue Tracker failure handling. - *No Wiki results* → APPLY SKILL FILE `phases/testgen-flow-data-collection.md` + the `data-collection` skill's Wiki failure handling. - *User declines / does not answer questions* → APPLY SKILL FILE `phases/testgen-flow-question-generation.md` `` "User explicitly declines to answer". - *Incomplete / missing requirements inputs* → APPLY SKILL FILE `phases/testgen-flow-requirements-document-generation.md` `` "Missing or empty inputs". - *Documentation search query example + ranking rule* → the `data-collection` skill's documentation-binding search/ranking behavior. - *Initial-prompt format examples* → APPLY SKILL FILE `phases/testgen-flow-project-config-loading.md`. - **Phase 6 80%-export-success threshold** → APPLY SKILL FILE `phases/testgen-flow-test-case-export.md` (`Threshold (80%) met` field + `PARTIAL — N/M exported` state). - **Phase 5 test-case-count guidance** → APPLY SKILL FILE `phases/testgen-flow-test-case-generation.md` ``. - **Model tiers** (phase `subagent_required_model`): `tier: complex` = heavy reasoning / multi-source synthesis / requirements engineering (Opus-class / GPT high-tier); `tier: workhorse` = structured execution / extraction / generation + export (Sonnet-class / GPT mid-tier). - APPLY SKILL FILE `phases/testgen-flow-project-config-loading.md` - Input: user request with an Issue Tracker ticket key/URL. Output: `plans/testgen-{TICKET-KEY}/initial-data.md`, project config file. - Required skills: `sensitive-data` (config / initial-data redaction pre-write gate) - Recommended skills: `questioning` - Update `testgen-state.md`; Phase 0 is not complete until its output spot-check passes. - APPLY SKILL FILE `phases/testgen-flow-data-collection.md` - Input: initial user request, initial-data.md. Output: `plans/testgen-{TICKET-KEY}/raw-data.md` with Issue Tracker + Wiki data. - Required skills: `data-collection` - Update `testgen-state.md`; Phase 1 is not complete until its output spot-check passes. - APPLY SKILL FILE `phases/testgen-flow-gap-and-contradiction-analysis.md` - Input: raw-data.md. Output: `plans/testgen-{TICKET-KEY}/analysis.md` with contradictions, gaps, ambiguities. - Required skills: `qa-knowledge` (`gap_analysis` mode) - Update `testgen-state.md`; Phase 2 is not complete until its output spot-check passes. - APPLY SKILL FILE `phases/testgen-flow-question-generation.md` - Input: analysis.md. Output: `plans/testgen-{TICKET-KEY}/questions.md`, `plans/testgen-{TICKET-KEY}/answers.md`. - **WAIT FOR USER** to fill answers in questions.md. Explicit approval required. Note, ask user to fill it in, not ask yourself. - Recommended skills: `questioning` - Update `testgen-state.md`; Phase 3 is not complete until its output spot-check passes. - APPLY SKILL FILE `phases/testgen-flow-requirements-document-generation.md` - Input: raw-data.md + analysis.md + answers.md. Output: `plans/testgen-{TICKET-KEY}/requirements.md`. - Present a summary and require explicit confirmation. - Required skills: `qa-knowledge` (`synthesis` mode) - Update `testgen-state.md`; Phase 4 is not complete until its output spot-check passes. - APPLY SKILL FILE `phases/testgen-flow-test-case-generation.md` - Input: requirements.md. Output: `plans/testgen-{TICKET-KEY}/test-scenarios.md` - **WAIT FOR USER** to review `test-scenarios.md` before Phase 6 export (phase-file gate, step 5.9) — present a summary and require explicit confirmation; per-phase confirmation per `` priority (3). - Required skills: `qa-knowledge` (`scenario_design` mode + config-resolved TMS FORMAT binding). - Update `testgen-state.md`; Phase 5 is not complete until its output spot-check passes. - `coding` is NOT used for the default manual-scenario output (writes stay under `plans/testgen-{TICKET-KEY}/`); apply it only if a write targets tracked repo files outside that folder, per ``. - APPLY SKILL FILE `phases/testgen-flow-test-case-export.md` - Input: test-scenarios.md. Output: test cases exported to Test Management System **and** a local export receipt at `plans/testgen-{TICKET-KEY}/export-report.md` (TMS IDs/URLs, per-case status, timestamp). The local receipt is the on-disk evidence Phase 6 ran successfully. - **WAIT FOR USER** to provide target location and confirm export. - Required skills: `qa-knowledge` (`scenario_design` mode + config-resolved TMS EXPORT binding). - Update `testgen-state.md`; Phase 6 is not complete until its output spot-check passes. - `coding` is NOT used for the default flow (TMS export + receipt under `plans/testgen-{TICKET-KEY}/`); apply it only if a write targets tracked repo files outside that folder (e.g. embedding TMS IDs into a version-controlled file), per ``. - **Skip-without-agreement / falsified-skip refusal** (this workflow owns the rule; subordinate to the `hitl` skill): a skip asserted but contradicted by `testgen-state.md` / disk evidence is refused — announce the specific missing state row / absent artifact, then start the earliest incomplete phase the same turn. - **Priority (highest never overridden → lowest):** (1) safety / destructive confirmations — incl. `` outside-output-dir confirmation; (2) Phase 3 + Phase 6 HITL gates (answer `questions.md` / confirm TMS target + export scope) — never skipped by user instruction; (3) per-phase user confirmation; (4) the verification-failure override below. - **Gate-type convention:** only the priority-(2) gates carry a `type="HITL"` attribute (Phases 3 + 6). The priority-(3) per-phase confirmations (Phases 0, 1, 2, 4, 5) are intentional user-pauses that deliberately carry **no** `type=` attribute — here `type=` marks the never-overridden gates, not every pause. - **Testgen binding for the override** (skip-verification gate only): the trigger is the user asserting a phase complete while `testgen-state.md` does not mark it AND the expected output is absent. Action — if `testgen-state.md` is missing, create it from the Phase 0 `` first; log a row into its `## Verification-Failure Overrides` (row format owned by that template); then start the earliest incomplete phase the same turn without invoking the `hitl` ask path. Uncertainty (partial state, ambiguous assertion) → fall back to the `hitl` ask. - Load failure for a required phase file or skill: retry once, stop, record in `testgen-state.md`, ask the user; never substitute silently. - Phases 5-6: apply `coding` whenever the phase writes any file outside `plans/testgen-{TICKET-KEY}/` — including "mixed outputs" (writes both inside and outside) or when repository edit scope was not explicitly confirmed in chat. When in doubt, apply. - Examples: apply for `cypress/e2e/login.spec.ts`; skip when writing only `plans/testgen-{TICKET-KEY}/test-scenarios.md`. `testgen-state.md` (sections `## Phase Completion Status`, `## Phase Details`, `## Metrics`, `## Verification-Failure Overrides`) and the per-ticket output-directory layout are **initialized and owned by Phase 0** (APPLY SKILL FILE `phases/testgen-flow-project-config-loading.md`) — see it for the canonical layout. Each subsequent phase updates the state file per `` and writes its output (paths in each phase block) under `plans/testgen-{TICKET-KEY}/`. Expected per-ticket artifact set (one validation can confirm all phases ran): `initial-data.md` + project config (Phase 0) · `raw-data.md` (1) · `analysis.md` (2) · `questions.md` + `answers.md` (3) · `requirements.md` (4) · `test-scenarios.md` (5) · `export-report.md` (6) · `testgen-state.md` (all). Full schema/layout owned by Phase 0. Subagents: `discoverer` · `architect` · `engineer`. Cross-phase skills: `qa-knowledge` (gap analysis, synthesis, scenario design + TMS bindings — loads its own assets at point of use); `data-collection` (Issue Tracker + Wiki collection); `questioning`; `sensitive-data` (redaction — Phase 0 config/initial-data pre-write gate; Phase 1 collection runs it via `data-collection`); `coding` (conditional — only for writes to tracked repo files outside `plans/testgen-{TICKET-KEY}/`, per ``). Integrations: Issue Tracker (ticket data extraction), Wiki (documentation retrieval), TMS (Phase 6 export), and additional documentation stores (e.g. Google Drive) when configured — per `` Terminology. - Sequential execution only: each phase builds on the previous - No assumptions: document all unknowns, ask user via HITL gates - Evidence-based: all requirements reference actual Issue Tracker / Wiki content - Traceability: link requirements to source and test cases to requirements - On completion: link `requirements.md` + exported cases back to the source ticket (attachment/comment) and archive `testgen-state.md` + outputs for traceability - Each phase has corresponding output file in output directory - State file reflects accurate phase completion status - HITL gates (Phase 3, 6) have explicit user approval evidence - Requirements trace back to Issue Tracker / Wiki sources - Test cases trace back to requirements - Skipping Phase 3 HITL gate leads to assumptions in requirements - Wiki child pages often contain critical detail — always check for children - TMS MCP may lack container creation — user may need to create target locations manually in TMS UI - Merging redundant test cases too aggressively can lose coverage