@rsaglobaltech/specgate 0.27.8 → 1.0.0-rc.1

package/dist/scripts/cli/commands/harness/RunCommand.js

@@ -62,6 +62,7 @@ exports.promptInFlight = promptInFlight;
 exports.setJsonMode = setJsonMode;
 exports.error = error;
 exports.parseArgs = parseArgs;
+exports.baselineProblem = baselineProblem;
 exports.isRegeneratedMatrix = isRegeneratedMatrix;
 exports.publishBranch = publishBranch;
 exports.printReport = printReport;
@@ -82,11 +83,19 @@ const prompt_1 = require("../../../harness/prompt");
 const agent_permission_1 = require("../../../lib/agent-permission");
 const init_1 = require("../../../harness/init");
 const in_flight_1 = require("../../../lib/in-flight");
+const approvals_1 = require("../../../lib/approvals");
+const role_packages_1 = require("../../../lib/role-packages");
+/**
+ * REQ-317: the role packages each requirement's prompts carried in this run,
+ * recorded on its result — the report counts them across runs.
+ */
+const packagesUsed = new Map();
 const harness_events_1 = require("../../../lib/harness-events");
 const clarification_1 = require("../../../lib/clarification");
 const screen_requirements_1 = require("../../../lib/screen-requirements");
 const harness_guard_1 = require("../../../lib/harness-guard");
 const quality_run_1 = require("../../../lib/quality-run");
+const harness_environment_1 = require("../../../lib/harness-environment");
 const HarnessEvents_1 = require("../../../../packages/core/src/domain/HarnessEvents");
 /**
  * `read` against the project as it will be after its open changes are
@@ -659,6 +668,17 @@ function runGate(worktreeDir, testCmd, timeoutMs, req = {}, settings = {}, repor
             ...(0, HarnessEvents_1.testCounts)(`${test.stdout || ""}${test.stderr || ""}`),
             ms: Date.now() - testStart,
         });
+        // REQ-279: a test command that could not reach Docker failed for the
+        // machine, not the code; the person starts Docker, the agent cannot.
+        const missing = test.status !== 0 ? (0, harness_environment_1.environmentFailure)(`${test.stdout || ""}${test.stderr || ""}`) : null;
+        if (missing) {
+            return {
+                ok: false,
+                stage: `test command: ${resolved}`,
+                output: `${test.stdout || ""}${test.stderr || ""}`,
+                couldNotRun: missing,
+            };
+        }
         if (test.status !== 0) {
             // Name the command. A gate that silently does the wrong thing — running
             // the whole suite because a filter did not apply, say — produces a
@@ -890,7 +910,10 @@ function agentChecks(worktreeDir, req, settings, timeoutMs, testCmd) {
                 said = runAgentOnce(worktreeDir, command, promptFor(`Review only; change no file. Compare the change below with ${req.requirement}, its ` +
                     "statement and every scenario. Answer with a first line `VERDICT: PASS` when the " +
                     "change does what the requirement says and nothing it forbids, or `VERDICT: FAIL` " +
-                    "followed by each reason as `- path:line — what is wrong`.\n\n```diff\n" +
+                    "followed by each reason as `- path:line — what is wrong`." +
+                    // REQ-318: one advice line per role package; it never decides the verdict.
+                    (0, role_packages_1.reviewAdvice)(judged(worktreeDir, (d) => (0, role_packages_1.packagesFor)(d, String(req.requirement || "")))) +
+                    "\n\n```diff\n" +
                     diff +
                     "\n```"), timeoutMs, "review");
             }
@@ -1258,7 +1281,7 @@ function attemptRequirement(req, ctx) {
         let agentMs = 0;
         /** Which roles ran this attempt, in order — the run record's answer to "who did this". */
         const profilesUsed = [];
-        const record = (stage) => {
+        const record = (stage, gateStage) => {
             event({
                 type: "attempt",
                 state: "end",
@@ -1271,6 +1294,8 @@ function attemptRequirement(req, ctx) {
                 agentMs,
                 totalMs: Date.now() - attemptStart,
                 profiles: [...profilesUsed],
+                // REQ-301: which stage rejected it is what says what to strengthen.
+                ...(gateStage ? { gateStage } : {}),
             });
             ledgerAttempts(attemptLog);
         };
@@ -1288,21 +1313,31 @@ function attemptRequirement(req, ctx) {
                 // implementing step that follows must see what it said. Building once
                 // up front meant findings only ever reached the *next* attempt, which
                 // is not what a reviewer is for.
-                const prompt = judged(worktreeDir, (dir) => (0, prompt_1.buildPrompt)(req, dir, {
-                    promptPrefix: settings.promptPrefix,
-                    hint,
-                    previousFailure: previousFailure || undefined,
-                    reviewFindings: step.advisory ? undefined : reviewFindings || undefined,
-                    attempt,
-                    maxAttempts: settings.maxAttempts,
-                    // Read from the worktree, which carries the matrix at the base
-                    // commit — the precedent must be work already accepted, not a
-                    // sibling requirement this same run happens to be writing.
-                    withPrecedents: settings.promptPrecedents === true,
-                    e2eDir: settings.e2eDir && (0, screen_requirements_1.onScreen)(worktreeDir, req.requirement)
-                        ? settings.e2eDir
-                        : undefined,
-                }));
+                let stepPackages = [];
+                const prompt = judged(worktreeDir, (dir) => {
+                    stepPackages = (0, role_packages_1.packagesFor)(dir, req.requirement).map((p) => p.id);
+                    return (0, prompt_1.buildPrompt)(req, dir, {
+                        promptPrefix: settings.promptPrefix,
+                        hint,
+                        previousFailure: previousFailure || undefined,
+                        reviewFindings: step.advisory ? undefined : reviewFindings || undefined,
+                        attempt,
+                        maxAttempts: settings.maxAttempts,
+                        // Read from the worktree, which carries the matrix at the base
+                        // commit — the precedent must be work already accepted, not a
+                        // sibling requirement this same run happens to be writing.
+                        withPrecedents: settings.promptPrecedents === true,
+                        e2eDir: settings.e2eDir && (0, screen_requirements_1.onScreen)(worktreeDir, req.requirement)
+                            ? settings.e2eDir
+                            : undefined,
+                    });
+                });
+                if (stepPackages.length > 0) {
+                    const seen = packagesUsed.get(req.requirement) || new Set();
+                    for (const id of stepPackages)
+                        seen.add(id);
+                    packagesUsed.set(req.requirement, seen);
+                }
                 const promptFile = path.join(os.tmpdir(), `specgate-harness-prompt-${req.requirement}-${crypto.randomBytes(4).toString("hex")}.md`);
                 fs.writeFileSync(promptFile, prompt, "utf8");
                 archivePrompt(worktreeDir, req.requirement, attempt, step, prompt);
@@ -1313,6 +1348,8 @@ function attemptRequirement(req, ctx) {
                     role: step.profile || (step.advisory ? "reviewer" : "agent"),
                     chars: prompt.length,
                     lines: prompt.split("\n").length,
+                    // REQ-317: which role packages it carried.
+                    ...(stepPackages.length > 0 ? { packages: stepPackages } : {}),
                 });
                 // D1: a profile chosen by matching this requirement, when one matches.
                 // The role ladder still wins — `attempt_profiles` names a profile for
@@ -1461,8 +1498,10 @@ function attemptRequirement(req, ctx) {
             warn(`${req.requirement}: ${gate.couldNotRun}`);
             for (const line of gate.output.trim().split("\n").slice(-5))
                 warn(`  ${line}`);
-            warn(`  not retrying: fix the check (harness.config.yaml) and run it again`);
-            record("check");
+            warn(gate.stage.startsWith("quality check")
+                ? `  not retrying: fix the check (harness.config.yaml) and run it again`
+                : `  not retrying: the agent cannot fix the machine`);
+            record("check", gate.stage);
             stopRetrying = true;
             break;
         }
@@ -1486,11 +1525,11 @@ function attemptRequirement(req, ctx) {
                         previousFailure;
                 warn(`${req.requirement}: the failure is in ${specFiles.join(", ")}, which the agent may not edit`);
                 warn(`  not retrying; fix the specification and run it again`);
-                record("spec");
+                record("spec", gate.stage);
                 stopRetrying = true;
                 break;
             }
-            record("gate");
+            record("gate", gate.stage);
             continue;
         }
         // Green — but does the diff contain what the row promised? (A2)
@@ -2053,6 +2092,9 @@ function processRequirement(req, ctx) {
         ledgerWrite();
     }
     const result = processOneRequirement(req, ctx);
+    const used = packagesUsed.get(req.requirement);
+    if (used && used.size > 0)
+        result.packages = [...used];
     if (ledger) {
         ledger.record.results.push(result);
         delete ledger.record.running;
@@ -2074,6 +2116,22 @@ function processOneRequirement(req, ctx) {
     });
     if (matchedProfile)
         info(`${req.requirement}: profile '${matchedProfile}' matched`);
+    // REQ-315: a requirement of an open change waits for its approvals — the 1.0
+    // default for a layered change — before any worktree or agent.
+    const owning = (0, in_flight_1.inFlightRequirements)(projectDir).find((r) => r.id === req.requirement);
+    const unapproved = owning ? (0, approvals_1.missingApprovals)(projectDir, owning.change) : null;
+    if (unapproved) {
+        error(`${req.requirement}: ${unapproved.message} [harness_not_approved]`);
+        warn(`  fix: ${unapproved.fix}`);
+        return {
+            requirement: req.requirement,
+            category: req.category,
+            result: "skipped",
+            attempts: 0,
+            branch,
+            error: `harness_not_approved: ${unapproved.message} Fix: ${unapproved.fix}`,
+        };
+    }
     const readiness = readinessOf(projectDir, req);
     const unrunnable = readiness.blockers.find((b) => b.code === "requirement_scenario_unrunnable");
     const hardStop = unrunnable || (ctx.skipNotReady && !readiness.ready);
@@ -2953,6 +3011,12 @@ class RunCommand extends command_1.BaseCommand {
             const identity = gitIdentityProblem(projectDir);
             if (identity)
                 throw new Error(identity);
+            // REQ-279: a stopped Docker fails every attempt in the tests or the
+            // checks, and the agent cannot start it (truck-flote S61: three attempts).
+            const needsDocker = (0, harness_environment_1.dockerNeed)(projectDir, settings.quality || null);
+            const noDocker = needsDocker ? (0, quality_run_1.dockerWhyNot)() : null;
+            if (noDocker)
+                throw new Error(`${noDocker} This project needs it: ${needsDocker}.`);
             if (args.chain) {
                 if (args.baseBranch) {
                     throw new Error("--chain merges into the branch you are on: check that branch out instead of --base-branch.");