run-uat

What It Does

run-uat is the slice-level acceptance testing agent in the auto-mode pipeline. It runs after a slice completes and before the next slice begins, giving the system a structured opportunity to verify that what was built actually meets the slice’s acceptance criteria. The prompt receives all relevant context pre-loaded — slice artifacts, implementation details, the UAT specification file, skill activations, and metadata about what type of test to run — so it can begin working immediately without re-reading planning files.

The prompt handles six distinct UAT modes, each with different automation rules:

• artifact-driven — verifies with shell commands, scripts, file reads, and artifact structure checks.
• browser-executable — uses browser tools to navigate to a target URL, verify expected behavior, and capture screenshots as evidence.
• runtime-executable — executes a specified command or script, captures stdout/stderr, and records pass/fail based on exit code and output.
• live-runtime — exercises the real runtime path by starting or connecting to the app/service and verifying observable behavior.
• mixed — runs all automatable artifact-driven and live-runtime checks, then separates any remaining human-only checks explicitly.
• human-experience — automates setup, preconditions, screenshots, and objective checks, but marks taste-based or experiential checks as NEEDS-HUMAN rather than inventing subjective PASS results.

For each check, the prompt selects the lightest evidence tool that honestly proves the check — shell commands, grep/ripgrep file searches, script invocations, file content reads, artifact structure verification, or browser/runtime/network inspection for live flows. When a check cannot be honestly automated, the best objective evidence is gathered and the check is marked NEEDS-HUMAN.

For each check the prompt records a description, the evidence mode used (artifact, runtime, or human-follow-up), the command or action taken, the actual result observed, and a per-check verdict of PASS, FAIL, or NEEDS-HUMAN.

After running all checks, an overall verdict is computed: PASS if all required checks passed, FAIL if any check failed, or PARTIAL if some checks were skipped, inconclusive, or still require human judgment. The result is always written to uatResultPath in a fixed schema — YAML frontmatter with sliceId, uatType, verdict, and date; a checks table; an overall verdict with summary; and a notes section capturing any additional context, errors encountered, screenshots or logs gathered, and manual follow-up still required. This structured format lets the dispatcher mechanically read the verdict and decide whether to advance to the next slice or pause for human review.

Pipeline Position

run-uat is dispatched by auto-mode as a post-slice validation step, not as part of the main task execution pipeline. It runs once per slice that has a UAT specification file, bridging complete-slice and the beginning of the next research-plan-execute cycle. A PARTIAL verdict — indicating NEEDS-HUMAN checks that cannot be mechanically verified — pauses auto-mode in the same way a FAIL does, surfacing the remaining checks for the user to perform manually.

Variables

Variable	Description	Required
milestoneId	Current milestone identifier	Yes
sliceId	Current slice identifier being tested	Yes
workingDirectory	Absolute path to the project working directory	Yes
inlinedContext	Pre-assembled context block containing slice artifacts and implementation details for UAT	Yes
skillActivation	Injected skill-loading instruction block; activates any skills that match the current UAT context	Yes
uatPath	File path to the UAT specification document	Yes
uatResultPath	File path where UAT results should be written	Yes
uatType	Automation mode for this UAT run — one of artifact-driven, browser-executable, runtime-executable, live-runtime, mixed, or human-experience	Yes

Used By

• /gsd auto — dispatched as a post-slice validation step when a UAT specification file exists for the completed slice