@entelligentsia/forgecli 0.7.9 → 0.7.10

package/dist/forge-payload/.tools/preflight-gate.cjs

@@ -0,0 +1,350 @@
+'use strict';
+
+// preflight-gate.cjs — evaluates a phase's declared gates against the current
+// task state, pre-spawn. Returns { ok, missing[] } so the orchestrator (or a
+// manual command) can halt loudly before invoking a subagent on broken
+// prerequisites.
+//
+// Pure function: only fs.existsSync / fs.statSync / fs.readFileSync. No writes,
+// no network, no process spawns.
+
+const fs = require('node:fs');
+const path = require('node:path');
+const { parseVerdict } = require('./parse-verdict.cjs');
+
+function preflight({ phase, gates, state = {}, substitutions = {}, verdictSources = {} }) {
+  const spec = gates && gates[phase];
+  if (!spec) {
+    return { ok: false, missing: [`no gate definition registered for phase "${phase}" (unknown phase or missing gates block)`] };
+  }
+
+  const missing = [];
+
+  for (const art of spec.artifacts || []) {
+    const resolved = applySubstitutions(art.path, substitutions);
+    let exists = false;
+    let size = 0;
+    try {
+      const st = fs.statSync(resolved);
+      exists = st.isFile();
+      size = st.size;
+    } catch (_) {
+      exists = false;
+    }
+    if (!exists) {
+      missing.push(`artifact missing: ${resolved}`);
+    } else if (size < (art.minBytes || 0)) {
+      missing.push(`artifact too small (stub): ${resolved} (${size} bytes, need >= ${art.minBytes})`);
+    }
+  }
+
+  for (const pred of spec.require || []) {
+    if (!evalPredicate(pred, state)) {
+      missing.push(`require failed: ${describePredicate(pred)} (got ${JSON.stringify(readField(pred.field, state))})`);
+    }
+  }
+
+  for (const pred of spec.forbid || []) {
+    if (evalPredicate(pred, state)) {
+      missing.push(`forbid triggered: ${describePredicate(pred)}`);
+    }
+  }
+
+  for (const after of spec.after || []) {
+    const src = verdictSources[after.phase];
+    if (!src) {
+      missing.push(`predecessor verdict source not provided for phase "${after.phase}"`);
+      continue;
+    }
+    let contents;
+    try {
+      contents = fs.readFileSync(src, 'utf8');
+    } catch (err) {
+      missing.push(`cannot read predecessor review for "${after.phase}" at ${src}: ${err.code || err.message}`);
+      continue;
+    }
+    const verdict = parseVerdict(contents);
+    if (verdict === null) {
+      missing.push(`predecessor review for "${after.phase}" has no parseable **Verdict:** line (${src})`);
+    } else if (verdict !== after.verdict) {
+      missing.push(`predecessor "${after.phase}" verdict is "${verdict}", expected "${after.verdict}"`);
+    }
+  }
+
+  return { ok: missing.length === 0, missing };
+}
+
+function applySubstitutions(template, subs) {
+  return template.replace(/\{(\w+)\}/g, (full, key) => {
+    if (Object.prototype.hasOwnProperty.call(subs, key)) return String(subs[key]);
+    return full;
+  });
+}
+
+function readField(dottedPath, state) {
+  const parts = dottedPath.split('.');
+  let cur = state;
+  for (const p of parts) {
+    if (cur === null || cur === undefined) return undefined;
+    cur = cur[p];
+  }
+  return cur;
+}
+
+function evalPredicate(pred, state) {
+  const actual = readField(pred.field, state);
+  switch (pred.op) {
+    case '==':
+      return String(actual) === String(pred.value);
+    case '!=':
+      return String(actual) !== String(pred.value);
+    case 'in':
+      return pred.value.map(String).includes(String(actual));
+    default:
+      throw new Error(`preflight-gate: unknown predicate op "${pred.op}"`);
+  }
+}
+
+function describePredicate(pred) {
+  if (pred.op === 'in') return `${pred.field} in [${pred.value.join(', ')}]`;
+  return `${pred.field} ${pred.op} ${pred.value}`;
+}
+
+// Canonical review artifact filenames per phase. Centralised here so the
+// orchestrator, manual commands, and tests all agree on where a given phase's
+// verdict lives.
+const VERDICT_ARTIFACTS = {
+  'review-plan': 'PLAN_REVIEW.md',
+  'review-code': 'CODE_REVIEW.md',
+  'validate':    'VALIDATION_REPORT.md',
+  'approve':     'ARCHITECT_APPROVAL.md',
+};
+
+module.exports = { preflight };
+
+// CLI shim: `node preflight-gate.cjs --phase <name> --task <taskId> [--bug <bugId>] [--workflow <name>]`
+// exit codes: 0 ok, 1 gate(s) failed, 2 invalid args / missing definitions
+// Scan the sprint directory for a subdirectory matching the task ID prefix.
+// Returns the directory name (e.g. "FORGE-S12-T06-model-discovery") or null.
+function resolveTaskArtifactDir(taskRecord, engineeringRoot) {
+  if (!taskRecord || !taskRecord.sprintId || !taskRecord.taskId) return null;
+  const sprintDir = path.resolve(process.cwd(), engineeringRoot, 'sprints', taskRecord.sprintId);
+  try {
+    const entries = fs.readdirSync(sprintDir);
+    for (const entry of entries) {
+      try {
+        if (fs.statSync(path.join(sprintDir, entry)).isDirectory() &&
+            entry.startsWith(taskRecord.taskId + '-')) {
+          return entry;
+        }
+      } catch (_) { /* skip unreadable entries */ }
+    }
+  } catch (_) { /* sprint directory not found */ }
+  return null;
+}
+
+if (require.main === module) {
+  const args = parseArgs(process.argv.slice(2));
+  if (!args.phase || (!args.task && !args.bug)) {
+    process.stderr.write('Usage: preflight-gate.cjs --phase <phaseName> --task <taskId> [--bug <bugId>]\n');
+    process.exit(2);
+  }
+
+  const { parseGates } = require('./parse-gates.cjs');
+  const store = require('./store.cjs');
+
+  // Resolve store records and substitutions BEFORE calling loadWorkflowMarkdown
+  // so the placeholder-key filter can use them to select the correct workflow file.
+  // (Previously loadWorkflowMarkdown was called first, causing fix_bug.md to shadow
+  // orchestrate_task.md for phases shared between the two workflows — forge#72.)
+  const taskRecord = args.task ? safe(() => store.getTask(args.task)) : null;
+  const bugRecord = args.bug ? safe(() => store.getBug(args.bug)) : null;
+  const state = {};
+  if (taskRecord) state.task = taskRecord;
+  if (bugRecord) state.bug = bugRecord;
+
+  // Resolve engineering root from config; path templates can reference {engineering}.
+  let engineeringRoot = 'engineering';
+  try {
+    const cfg = JSON.parse(fs.readFileSync(path.resolve(process.cwd(), '.forge/config.json'), 'utf8'));
+    if (cfg.paths && cfg.paths.engineering) engineeringRoot = cfg.paths.engineering;
+  } catch (_) { /* fall back to default */ }
+
+  // {task} / {bug} in path templates refer to the artifact directory suffix
+  // (e.g., "FORGE-S12-T06-model-discovery", "BUG-007-broken-foo").
+  // task.path is the primary source file in forge/, NOT the artifact directory.
+  // Scan the sprint directory to find the correct artifact directory name.
+  function lastSegment(p) {
+    const parts = String(p || '').split('/').filter(Boolean);
+    return parts[parts.length - 1] || '';
+  }
+  const taskArtifactDir = resolveTaskArtifactDir(taskRecord, engineeringRoot);
+  const taskDir = taskArtifactDir
+    || (taskRecord && taskRecord.path ? lastSegment(taskRecord.path) : args.task);
+  const bugDir = bugRecord && bugRecord.path ? lastSegment(bugRecord.path) : args.bug;
+
+  // Compute the full artifact directory path for verdict source resolution.
+  let taskArtifactPath = null;
+  if (taskArtifactDir && taskRecord && taskRecord.sprintId) {
+    taskArtifactPath = path.join(engineeringRoot, 'sprints', taskRecord.sprintId, taskArtifactDir);
+  } else if (taskRecord && taskRecord.path) {
+    taskArtifactPath = taskRecord.path;
+  }
+
+  const substitutions = {
+    engineering: engineeringRoot,
+    sprint: taskRecord ? taskRecord.sprintId : undefined,
+    task: taskDir,
+    bug: bugDir,
+  };
+
+  // Now load the workflow, passing substitutions so the placeholder-key filter
+  // can skip workflows whose gate block references keys not present in subs.
+  const workflowMd = loadWorkflowMarkdown(args.phase, args.workflow, substitutions);
+  if (!workflowMd) {
+    process.stderr.write(`preflight-gate: could not locate workflow file defining phase "${args.phase}"\n`);
+    process.exit(2);
+  }
+  let gates;
+  try {
+    gates = parseGates(workflowMd);
+  } catch (err) {
+    process.stderr.write(`preflight-gate: ${err.message}\n`);
+    process.exit(2);
+  }
+  if (!gates[args.phase]) {
+    process.stderr.write(`preflight-gate: no gates block for phase "${args.phase}"\n`);
+    process.exit(2);
+  }
+
+  const verdictSources = resolveVerdictSources(gates[args.phase].after || [], taskArtifactPath, bugRecord);
+
+  const result = preflight({ phase: args.phase, gates, state, substitutions, verdictSources });
+  if (result.ok) process.exit(0);
+  process.stderr.write(`Gate failed for phase "${args.phase}":\n`);
+  for (const m of result.missing) process.stderr.write(`  - ${m}\n`);
+  process.exit(1);
+}
+
+function parseArgs(argv) {
+  const out = {};
+  for (let i = 0; i < argv.length; i++) {
+    const a = argv[i];
+    if (a === '--phase') out.phase = argv[++i];
+    else if (a === '--task') out.task = argv[++i];
+    else if (a === '--bug') out.bug = argv[++i];
+    else if (a === '--workflow') out.workflow = argv[++i];
+  }
+  return out;
+}
+
+function safe(fn) {
+  try { return fn(); } catch (_) { return null; }
+}
+
+// Extract the gate block body for a given phase from a workflow markdown string.
+// Returns the text between the opening and closing fence for that phase, or null.
+function extractGateBlockBody(md, phaseName) {
+  const openPattern = new RegExp('^```gates\\s+phase=' + escapeRegex(phaseName) + '\\s*$', 'm');
+  const openMatch = openPattern.exec(md);
+  if (!openMatch) return null;
+  const afterOpen = md.slice(openMatch.index + openMatch[0].length);
+  const closeIdx = afterOpen.indexOf('\n```');
+  if (closeIdx === -1) return afterOpen; // unterminated fence — return what we have
+  return afterOpen.slice(0, closeIdx);
+}
+
+// Check whether all {placeholder} keys in a gate block body are satisfied by
+// the given substitutions map. Returns true if all placeholders are satisfied
+// (or there are none), false if any placeholder key is missing from subs.
+function gatePlaceholdersSatisfied(gateBody, subs) {
+  const tokenRe = /\{(\w+)\}/g;
+  let match;
+  while ((match = tokenRe.exec(gateBody)) !== null) {
+    const key = match[1];
+    if (!Object.prototype.hasOwnProperty.call(subs, key) || subs[key] === undefined) {
+      return false;
+    }
+  }
+  return true;
+}
+
+function loadWorkflowMarkdown(phaseName, workflowName, substitutions) {
+  const workflowsDir = path.resolve(process.cwd(), '.forge/workflows');
+  let entries;
+  try {
+    entries = fs.readdirSync(workflowsDir).filter((f) => f.endsWith('.md'));
+  } catch (_) {
+    return null;
+  }
+  const fencePattern = new RegExp('^```gates\\s+phase=' + escapeRegex(phaseName) + '\\s*$', 'm');
+
+  // If a specific workflow file was requested, try it first before scanning all files.
+  // This prevents alphabetically-earlier files from shadowing the caller's workflow.
+  if (workflowName) {
+    const normalised = workflowName.endsWith('.md') ? workflowName : workflowName + '.md';
+    if (entries.includes(normalised)) {
+      const md = fs.readFileSync(path.join(workflowsDir, normalised), 'utf8');
+      if (fencePattern.test(md)) return md;
+    }
+  }
+
+  // Placeholder-key filter: when substitutions are provided, skip any workflow
+  // whose gate block for this phase references a {key} not present in subs.
+  // This prevents fix_bug.md (which uses {bug}) from shadowing orchestrate_task.md
+  // (which uses {task}) when only --task is supplied and --bug is absent.
+  // Fall-back: if NO workflow passes the filter, return the first match regardless
+  // (preserves existing behaviour for malformed/unknown invocations).
+  const subs = substitutions || {};
+  const hasSubstitutions = Object.keys(subs).some(k => subs[k] !== undefined);
+
+  let firstMatch = null; // fallback candidate
+  for (const entry of entries) {
+    const md = fs.readFileSync(path.join(workflowsDir, entry), 'utf8');
+    if (!fencePattern.test(md)) continue;
+
+    if (firstMatch === null) firstMatch = md; // remember first match for fallback
+
+    if (!hasSubstitutions) {
+      // No substitutions provided — use original first-match behaviour
+      return md;
+    }
+
+    const gateBody = extractGateBlockBody(md, phaseName);
+    if (gateBody !== null && gatePlaceholdersSatisfied(gateBody, subs)) {
+      return md;
+    }
+    // Placeholder(s) unsatisfied — log at warn level and continue scanning
+    process.stderr.write(
+      `preflight-gate: skipping ${entry} for phase "${phaseName}" — gate block contains ` +
+      `placeholder key(s) not present in supplied substitutions\n`
+    );
+  }
+
+  // Fallback: no workflow passed the placeholder filter — return first match to
+  // avoid total breakage for malformed invocations (preserves existing behaviour)
+  if (firstMatch !== null) {
+    process.stderr.write(
+      `preflight-gate: placeholder filter matched no workflow for phase "${phaseName}" — ` +
+      `falling back to first match\n`
+    );
+    return firstMatch;
+  }
+  return null;
+}
+
+function escapeRegex(s) {
+  return s.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
+}
+
+function resolveVerdictSources(afterList, taskArtifactPath, bugRecord) {
+  const sources = {};
+  const base = taskArtifactPath || (bugRecord ? bugRecord.path : null);
+  if (!base) return sources;
+  for (const entry of afterList) {
+    const filename = VERDICT_ARTIFACTS[entry.phase];
+    if (!filename) continue; // unknown predecessor phase — preflight will flag it via missing-source
+    sources[entry.phase] = path.resolve(process.cwd(), base, filename);
+  }
+  return sources;
+}

package/dist/forge-payload/.tools/prompts/sprint-plan-prompt.md

@@ -0,0 +1,70 @@
+# Sprint Plan — Architect Persona Prompt
+
+You are acting as the **Forge Architect** (🗻).
+
+Your role: decompose a sprint's requirements into a concrete, dependency-ordered task list. You are a senior technical architect who thinks carefully about sequencing, blast radius, and iron laws before writing a single line of output.
+
+---
+
+## Your Task
+
+Read the sprint requirements provided and produce a **JSON array** of task objects. Each task covers one coherent, implementable unit of work. Tasks must be small enough to be planned and implemented in a single sprint sub-task cycle.
+
+---
+
+## Output Format
+
+Emit **only** a JSON array — no preamble, no markdown fences, no commentary outside the JSON. The schema:
+
+```json
+[
+  {
+    "taskId": "SPRINT_ID-T01",
+    "title": "Short imperative title (≤80 chars)",
+    "estimate": "S",
+    "dependencies": [],
+    "pipeline": "plan,implement,review,validate,approve,commit",
+    "acceptanceCriteria": [
+      "Specific, verifiable criterion 1",
+      "Specific, verifiable criterion 2"
+    ]
+  }
+]
+```
+
+### Field rules
+
+| Field | Type | Rule |
+|---|---|---|
+| `taskId` | string | `{SPRINT_ID}-T{NN}` — sequential integers starting at 01. Use the sprint ID from the requirements. |
+| `title` | string | Imperative phrase, ≤80 chars. No trailing punctuation. |
+| `estimate` | enum | One of: `S` (≤4h), `M` (≤1d), `L` (≤2d), `XL` (≤5d). |
+| `dependencies` | string[] | Task IDs of tasks this task depends on. Empty array if none. No circular deps. |
+| `pipeline` | string | Default: `"plan,implement,review,validate,approve,commit"`. Adjust only for trivial chores. |
+| `acceptanceCriteria` | string[] | 2–8 specific, verifiable criteria. Each criterion must be independently checkable. No vague phrases like "works correctly". |
+
+### Dependency rules
+
+- Dependencies form a DAG (directed acyclic graph). Circular dependencies are invalid.
+- A release/packaging task always depends on all feature tasks.
+- Tasks that share no code boundary may be declared parallel (no mutual dep).
+- Order tasks so lower-numbered tasks are prerequisites for higher-numbered tasks where possible.
+
+---
+
+## Quality Checklist
+
+Before emitting output, verify:
+
+- [ ] Every `taskId` is unique and follows the `{SPRINT_ID}-T{NN}` pattern
+- [ ] No circular dependencies
+- [ ] Each task has at least 2 acceptance criteria
+- [ ] Estimates are realistic (L tasks should not attempt to cover more than 2 full days of work)
+- [ ] A release or packaging task (if applicable) is listed last and depends on all feature tasks
+- [ ] The JSON is valid — parseable without error
+
+---
+
+## Sprint Requirements
+
+{SPRINT_REQUIREMENTS}

package/dist/forge-payload/.tools/schemas/task-list.schema.json

@@ -0,0 +1,53 @@
+{
+  "$schema": "http://json-schema.org/draft-07/schema#",
+  "title": "TaskList",
+  "description": "JSON Schema for the LLM-generated task list produced by forge:sprint-plan.",
+  "type": "array",
+  "minItems": 1,
+  "items": {
+    "type": "object",
+    "required": ["taskId", "title", "estimate", "dependencies", "pipeline", "acceptanceCriteria"],
+    "additionalProperties": false,
+    "properties": {
+      "taskId": {
+        "type": "string",
+        "pattern": "^[A-Z][A-Z0-9_-]+-T[0-9]+$",
+        "description": "Task identifier in the form {SPRINT_ID}-T{NN}, e.g. FORGE-S20-T01"
+      },
+      "title": {
+        "type": "string",
+        "minLength": 1,
+        "maxLength": 80,
+        "description": "Short imperative title for the task"
+      },
+      "estimate": {
+        "type": "string",
+        "enum": ["S", "M", "L", "XL"],
+        "description": "Size estimate: S (≤4h), M (≤1d), L (≤2d), XL (≤5d)"
+      },
+      "dependencies": {
+        "type": "array",
+        "items": {
+          "type": "string",
+          "pattern": "^[A-Z][A-Z0-9_-]+-T[0-9]+$"
+        },
+        "description": "Task IDs this task depends on. Empty array if no dependencies."
+      },
+      "pipeline": {
+        "type": "string",
+        "minLength": 1,
+        "description": "Comma-separated pipeline phase list, e.g. plan,implement,review,validate,approve,commit"
+      },
+      "acceptanceCriteria": {
+        "type": "array",
+        "minItems": 2,
+        "maxItems": 12,
+        "items": {
+          "type": "string",
+          "minLength": 10
+        },
+        "description": "Specific, verifiable acceptance criteria (2–12 items)"
+      }
+    }
+  }
+}