issue-flow 0.12.0 → 0.14.0

package/README.md

@@ -1,310 +1,115 @@
 # issue-flow
 
-> This is the npm package README. For full project documentation, see the [root README](../../README.md).
+**Turn an issue into a reviewed Pull Request, without sitting in front of it.**
 
-Unified CLI that orchestrates the full issue-to-PR pipeline via [Claude Code](https://docs.anthropic.com/en/docs/claude-code) Headless mode. Analyzes issues, generates PRDs, creates task plans, implements code iteratively, reviews results, and opens pull requests -- all programmatically, without interactive sessions.
-
-Built on the [Ralph pattern](https://ghuntley.com/ralph/) for autonomous AI agent loops.
-
-## Pipeline Flow
-
-```mermaid
-flowchart LR
-    subgraph "issue-flow run 42"
-        A["init"] --> B["analyze"]
-        B --> C["prd"]
-        C --> D["plan"]
-        D --> E["execute"]
-        E --> F["review"]
-        F --> G{PASS?}
-        G -- Yes --> H["pr"]
-        G -- No --> I{"Retries\n< max?"}
-        I -- Yes --> E
-        I -- No --> J["Stop"]
-    end
-```
-
-## Requirements
-
-- **Node.js** >= 18.0.0
-- **git** installed and available in PATH
-- **Claude Code** (`npm install -g @anthropic-ai/claude-code`)
-- **GitHub CLI** (`gh`) authenticated (`gh auth login`)
-
-Run `npx issue-flow init` to verify all prerequisites.
-
-## Installation
+A CLI that orchestrates the whole path — analyse, plan, implement, verify,
+review, deliver — by driving a coding agent in headless mode:
+[Claude Code](https://docs.anthropic.com/en/docs/claude-code) (the default),
+[Codex CLI](https://developers.openai.com/codex/noninteractive), Cursor CLI or
+[Antigravity CLI](https://antigravity.google/docs/cli/getting-started/), one
+agent per phase if you want.
 
 ```bash
-# Run directly via npx (no install needed)
-npx issue-flow run 42
-
-# Or install globally
-npm install -g issue-flow
-issue-flow run 42
+npx issue-flow init           # check prerequisites and repository conventions
+npx issue-flow run 42         # prd → plan → execute → review → pr
+npx issue-flow run 42 --web   # …and watch it live at http://localhost:3737
 ```
 
-## Global Options
-
-All commands support the following options:
-
-| Flag | Description |
-|------|-------------|
-| `-v, --verbose` | Show Claude progress output in real time |
-| `-t, --timeout <seconds>` | Override headless timeout in seconds (0 = no limit) |
+> **Full documentation lives in the repository:**
+> [README](https://github.com/fabioassuncao/issue-flow#readme) ·
+> [Commands](https://github.com/fabioassuncao/issue-flow/blob/main/docs/commands.md) ·
+> [Configuration](https://github.com/fabioassuncao/issue-flow/blob/main/docs/configuration.md) ·
+> [Agents](https://github.com/fabioassuncao/issue-flow/blob/main/docs/agents.md) ·
+> [Storage](https://github.com/fabioassuncao/issue-flow/blob/main/docs/storage.md) ·
+> [Web monitoring](https://github.com/fabioassuncao/issue-flow/blob/main/docs/web-monitor.md) ·
+> [Resilience](https://github.com/fabioassuncao/issue-flow/blob/main/docs/resilience.md)
 
-```bash
-# Run with verbose output
-npx issue-flow analyze 42 --verbose
-
-# Override timeout to 10 minutes
-npx issue-flow analyze 42 --timeout 600
-
-# Disable timeout entirely
-npx issue-flow run 42 --timeout 0
-
-# Combine flags
-npx issue-flow analyze 42 -v -t 600
-```
+## Requirements
 
-Every phase that invokes Claude once (`analyze`, `prd`, `plan`, `review`, `pr`, `pr-review`, `generate`) shares a default limit of **15 minutes** per invocation; `execute` has no limit, since its iteration budget is what bounds it. Use `--timeout` to raise it on a large issue, or `--timeout 0` to remove it. When an invocation is cut short the phase reports it as a timeout and retries with backoff, so the limit is the first thing to raise if a phase keeps dying at the same elapsed time.
+- **Node.js** ≥ 22
+- **Git**, available in `PATH`, inside a repository
+- **A coding agent** — `npm install -g @anthropic-ai/claude-code` for the default
+- **GitHub CLI** (`gh`), authenticated — only for GitHub issues; a run on local
+  issues does not need it
+
+## What it does
+
+- The full pipeline `prd` → `plan` → `execute` → `review` → `pr`, plus an
+  optional whole-PR review. Every phase is also a standalone command.
+- An iterative execute loop: each iteration is a fresh agent instance that picks
+  the highest-priority pending user story, implements it, runs quality checks and
+  commits.
+- An objective acceptance contract (typecheck, lint, tests) before the LLM
+  judges. An empty contract finishes `unverified`, never green.
+- Per-phase agent selection, resolved explicitly and never inferred from which
+  binary happens to be installed.
+- Resilience for long unattended runs: a failure taxonomy, per-kind retry
+  budgets, provider failover with circuit breakers, an inactivity watchdog and an
+  append-only event journal.
+- Multi-issue queues discovered from sub-issues and dependencies: one branch, one
+  Pull Request.
+- A read-only live web monitor, one card per active run across every project.
+- Issues from GitHub or from plain files — the local provider needs nothing
+  beyond git.
+- Repository conventions (templates, labels, base branch, commit and branch
+  format) discovered rather than imposed.
 
 ## Commands
 
-### `run` -- Full pipeline (end-to-end)
-
-```bash
-# Run the complete pipeline for an issue
-npx issue-flow run 42
-
-# Resume from a specific phase
-npx issue-flow run 42 --from execute
-
-# Manual mode (artifacts only, no execution)
-npx issue-flow run 42 --mode manual
-
-# Watch the run live in the browser (see "Web Monitoring" below)
-npx issue-flow run 42 --web
+| Command | What it does |
+|---------|--------------|
+| `run <issues...>` | The full pipeline, for one issue or a queue |
+| `resume [issue]` | Continue an interrupted pipeline, explicitly |
+| `generate` | Draft and create an issue on GitHub, locally, or both |
+| `init` | Check prerequisites and report (or create) missing conventions |
+| `analyze`, `prd`, `plan`, `execute`, `review`, `pr`, `pr-review` | The phases, standalone |
+| `status`, `ps`, `runs`, `logs`, `usage`, `pause`, `cancel` | Operate a running pipeline |
+| `agent`, `policy`, `conventions`, `routing` | Inspect what was resolved, and why |
+| `web serve`, `web stop` | The monitoring server |
 
-# Continue User Story numbering from the last used in this project
-npx issue-flow run 42 --continue
+Run `issue-flow <command> --help` for the flags, or read the
+[command reference](https://github.com/fabioassuncao/issue-flow/blob/main/docs/commands.md).
 
-# Force User Story numbering to start at a specific number
-npx issue-flow run 42 --start-us 27
-```
-
-Executes all phases in order: **init** → **analyze** → **prd** → **plan** → **execute** → **review** → **pr**. Automatically resumes from the last incomplete phase if pipeline state exists. On review failure, its findings are saved verbatim to `lastReviewFindings` in `tasks.json`, and a correction cycle (re-execute + re-review) runs up to `maxCorrectionCycles`. The re-execute step reads `lastReviewFindings` and treats the issue as unresolved even if every user story already has `passes: true`, until the findings are addressed and the field is cleared back to `null`.
-
-`--continue` and `--start-us <n>` control User Story (`US-NNN`) numbering continuity across `plan` runs of the same project — see [`plan`](#plan----convert-prd-to-task-plan) below.
-
-### `init` -- Check prerequisites
-
-```bash
-npx issue-flow init
-```
-
-Verifies that `claude`, `gh` (authenticated), and `git` (inside a repo) are available. Reports pass/fail for each with install hints.
+## Where things are written
 
-### `analyze` -- Analyze an issue
-
-```bash
-npx issue-flow analyze 42
-```
-
-Invokes Claude headlessly to fetch issue data, analyze the codebase, and produce a structured analysis saved to `~/.issue-flow/…/issues/42/analysis.md` (see [Pipeline State](#pipeline-state)).
-
-### `prd` -- Generate a PRD
-
-```bash
-npx issue-flow prd 42
-```
-
-Generates a Product Requirements Document from the issue analysis. Reads `analysis.md` from the same directory as context if available. Saves to `~/.issue-flow/…/issues/42/prd.md`.
-
-### `plan` -- Convert PRD to task plan
-
-```bash
-npx issue-flow plan 42
-
-# Continue User Story numbering from the last used in this project
-npx issue-flow plan 42 --continue
-
-# Force numbering to start at a specific number, ignoring history
-npx issue-flow plan 42 --start-us 27
-```
-
-Converts the PRD into a structured `~/.issue-flow/…/issues/42/tasks.json` with ordered user stories, acceptance criteria, and pipeline state. Validates the output with zod schemas.
-
-`US-NNN` numbering continues automatically across `plan` runs of the same project: the highest number already used in any of the project's `tasks.json` files is recovered and the next story continues from there, falling back to `US-001` when the project has no history yet. `--continue` names that automatic recovery explicitly; `--start-us <n>` forces a specific starting number instead, ignoring history — the two flags are mutually exclusive. The decision is always printed to the terminal and recorded in the project's `metadata.json` (`userStoryNumbering`) for audit. See the root [README](https://github.com/fabioassuncao/issue-flow#user-story-numbering-continuity) for the full cascade.
-
-### `execute` -- Run the story execution loop
-
-```bash
-npx issue-flow execute --issue 42
-npx issue-flow execute --issue 42 --max-iterations 15
-npx issue-flow execute --issue 42 --retry-forever
-```
-
-Runs the iterative agent loop. Each iteration is a fresh Claude instance that picks the next pending story, implements it, runs quality checks, and commits.
-
-| Flag | Description |
-|------|-------------|
-| `--issue N` | Issue number -- reads artifacts from `~/.issue-flow/…/issues/N/` |
-| `--max-iterations N` | Stop after N iterations (default: unlimited) |
-| `--retry-limit N` | Retry transient Claude failures up to N consecutive times (default: 10) |
-| `--retry-forever` | Retry transient Claude failures indefinitely |
-
-### `review` -- Validate the implementation
-
-```bash
-npx issue-flow review 42
-```
-
-Invokes Claude headlessly to verify acceptance criteria, run tests, and check for regressions. Outputs `PASS` or `FAIL` with findings.
-
-### `pr` -- Create a pull request
-
-```bash
-npx issue-flow pr 42
-```
-
-Creates a well-structured PR referencing the issue, with summary and test plan.
-
-### `generate` -- Create a new issue
-
-```bash
-npx issue-flow generate --prompt "Add dark mode support to the settings page"
-```
-
-Analyzes the project and creates a detailed GitHub issue via Claude headless.
-
-## Web Monitoring
-
-`run` and `execute` support an optional (off by default) real-time monitoring mode: a local HTTP server (plain `node:http`, zero new dependencies) serves a self-contained, read-only web UI with live progress -- current phase and activity, user stories, commits, pull requests, logs, and time estimates.
-
-```bash
-# Enable with defaults -- binds to 0.0.0.0:3737, reachable from your LAN/VPN
-npx issue-flow run 42 --web
-
-# Restrict to this machine only
-npx issue-flow run 42 --web --host 127.0.0.1
-```
-
-Each setting resolves with the precedence **CLI flag > environment variable > `.issue-flow.json` > default**:
-
-| CLI flag | Environment variable | `.issue-flow.json` key | Default |
-|----------|----------------------|------------------------|---------|
-| `--web` / `--serve` | `ISSUE_FLOW_WEB` | `web.enabled` | `false` |
-| `--port <n>` | `ISSUE_FLOW_WEB_PORT` | `web.port` | `3737` |
-| `--host <h>` | `ISSUE_FLOW_WEB_HOST` | `web.host` | `0.0.0.0` |
-| `--refresh <s>` | `ISSUE_FLOW_WEB_REFRESH` | `web.refreshSeconds` | `5` |
-| `--web-log-limit <n>` | `ISSUE_FLOW_WEB_LOG_LIMIT` | `web.logLimit` | `200` |
-| `--web-no-logs` | -- | `web.includeLogs` | logs included |
-
-Monitoring never affects the pipeline: with `--web` off the behavior is byte-for-byte identical, a busy port just skips the server with a warning, and killing the server mid-run has no effect on the execution. While enabled, the snapshot served at `/api/status` is also persisted to `~/.issue-flow/…/issues/N/session.json` -- outside your working tree, so there is nothing to add to `.gitignore`.
-
-For the full documentation (endpoints, `session.json` format, Tailscale setup), see the [root README](../../README.md#web-monitoring).
-
-## Pipeline State
-
-Each issue's state is tracked in a directory of its own under `~/.issue-flow`, keyed by a deterministic project id and the issue identifier:
+Nothing is written inside your repository. Artifacts live in a machine-wide tree
+keyed by a deterministic project id:
 
 ```
 ~/.issue-flow/projects/<project-id>/issues/42/
-  analysis.md    # Issue analysis
-  prd.md         # Product requirements
-  tasks.json     # Task plan with pipeline state and user stories
-  progress.txt   # Execution log
-  session.json   # Live session snapshot (web monitoring only)
-  pr-review/     # PR review reports and index
+  prd.md   tasks.json   progress.txt   session.json   events.jsonl   pr-review/
 ```
 
-Nothing is written to your repository. A `<projectRoot>/issues/` tree left by an earlier release is copied here automatically on first use and then treated as read-only -- see the [root README](../../README.md#global-storage) for the full layout, the project id derivation and the migration.
-
-The `pipeline` field tracks which phases have completed, enabling resume from any point:
-
-```json
-{
-  "pipeline": {
-    "analyzeCompleted": true,
-    "prdCompleted": true,
-    "jsonCompleted": true,
-    "executionCompleted": false,
-    "reviewCompleted": false,
-    "prCreated": false
-  }
-}
-```
+`ISSUE_FLOW_HOME` relocates the whole tree. A legacy `<projectRoot>/issues/`
+directory from an earlier release is copied in automatically on first use and
+then left read-only.
 
-The top-level `lastReviewFindings` field (`string | null`) holds the verbatim findings of the most recent failed `review` phase. Non-null overrides the "issue already complete" check even when every user story has `passes: true`, so a correction cycle's re-execute step is guaranteed to run instead of exiting immediately — see `core/engine.ts`.
+## Configuration
 
-## Architecture
-
-```
-src/
-  cli.ts                  # Entry point, subcommand registration (commander)
-  config.ts               # Configuration resolution and defaults
-  types.ts                # Shared TypeScript interfaces
-  schemas.ts              # Zod validation schemas
-  commands/
-    init.ts               # Prerequisite verification
-    generate.ts           # Headless issue creation
-    run.ts                # Full pipeline orchestrator
-    analyze.ts            # Headless issue analysis
-    prd.ts                # Headless PRD generation
-    plan.ts               # Headless PRD-to-JSON conversion
-    execute.ts            # Iterative story execution (engine wrapper)
-    review.ts             # Headless implementation review
-    pr.ts                 # Headless PR creation
-  core/
-    engine.ts             # Main agent loop
-    executor.ts           # Claude CLI invocation via execa
-    headless.ts           # Typed wrapper for claude -p invocations
-    pipeline.ts           # Pipeline state machine
-    state-manager.ts      # Typed CRUD for tasks.json
-    prompt-resolver.ts    # Prompt resolution and templating
-    session-state.ts      # Session snapshot reducer and publishers
-    session-publisher.ts  # Global session publisher slot
-    session-git.ts        # Commit/PR enrichment for the snapshot
-  web/
-    server.ts             # Web monitoring HTTP server (assets in web/public/)
-  ui/
-    logger.ts             # Colored logging utilities
-    progress.ts           # Progress bar and iteration headers
-    summary.ts            # Box drawing and summary display
-  utils/
-    shell.ts              # Shell command execution
-    git.ts                # Git operations
-    retry.ts              # Transient failure detection and backoff
-```
+Everything resolves through **CLI flag > environment variable >
+`.issue-flow.json` > `~/.issue-flow/config.json` > default**. Nothing is
+mandatory. See the
+[configuration reference](https://github.com/fabioassuncao/issue-flow/blob/main/docs/configuration.md).
 
 ## Development
 
 ```bash
-# Install dependencies
 npm install
-
-# Build
-npm run build
-
-# Type check
+npm run build       # tsup → dist/
 npm run typecheck
-
-# Run tests
 npm test
-
-# Watch mode
-npm run dev
+npm run smoke       # end-to-end, against deterministic stand-ins for claude and gh
+npm run check       # biome + typecheck
 ```
 
-For the full development setup and local testing guide, see [CONTRIBUTING.md](CONTRIBUTING.md).
-The official release procedure (changelog, version bump, tag, `npm publish`, GitHub
-Release) is documented in [CONTRIBUTING.md → Release process](CONTRIBUTING.md#release-process).
+See [CONTRIBUTING.md](CONTRIBUTING.md) for the full setup and the release
+process.
 
 ## Credits
 
-Based on [Geoffrey Huntley's Ralph pattern](https://ghuntley.com/ralph/) and the [snarktank/ralph](https://github.com/snarktank/ralph) repository.
+Based on [Geoffrey Huntley's Ralph pattern](https://ghuntley.com/ralph/) and the
+[snarktank/ralph](https://github.com/snarktank/ralph) repository.
 
-## See Also
+## License
 
-- [Skills & Sub-Agent Architecture](../../docs/skills-and-agents.md) -- Using Issue Flow interactively via Claude Code skills and the `resolve-issue` sub-agent.
+[MIT](LICENSE)

package/dist/agent-OX3KPHE4.js

@@ -0,0 +1,30 @@
+#!/usr/bin/env node
+import {
+  AGENT_COMMAND_SCHEMA_VERSION,
+  persistFirstAgentChoice,
+  runAgent,
+  runAgentUse,
+  writeAgentPreference
+} from "./chunk-UJ3L2MS7.js";
+import "./chunk-QJ5UZAH5.js";
+import "./chunk-QK7WUO7A.js";
+import "./chunk-TYRNTSPJ.js";
+import "./chunk-6QYMSNVA.js";
+import "./chunk-3WFZETIM.js";
+import "./chunk-HAKDLR6N.js";
+import "./chunk-BEZ5BA5U.js";
+import "./chunk-67YIU2VB.js";
+import "./chunk-P3LRUNC6.js";
+import "./chunk-G5EK2XRN.js";
+import "./chunk-J3SOTUVI.js";
+import "./chunk-LBGHI4S2.js";
+import "./chunk-7Z4Y3VKM.js";
+import "./chunk-XEYH6ZZI.js";
+export {
+  AGENT_COMMAND_SCHEMA_VERSION,
+  persistFirstAgentChoice,
+  runAgent,
+  runAgentUse,
+  writeAgentPreference
+};
+//# sourceMappingURL=agent-OX3KPHE4.js.map

package/dist/{analyze-OEET7KGE.js → analyze-WIND2SEB.js}

@@ -2,36 +2,49 @@
 import {
   issuePlaceholders,
   resolveCommandIssue
-} from "./chunk-73ATPIYX.js";
-import {
-  DEFAULT_HEADLESS_TIMEOUT_MS,
-  runHeadless
-} from "./chunk-4JSOYE4G.js";
+} from "./chunk-REO3PXXO.js";
+import "./chunk-PZTWU277.js";
 import "./chunk-KMMASOA6.js";
 import {
   publishPhaseMetrics,
   resolvePolicyPlaceholders
-} from "./chunk-MJULIKQU.js";
+} from "./chunk-PHW2YV6N.js";
 import {
   applyPlaceholders,
   loadPrompt
-} from "./chunk-LAFNLLQP.js";
-import "./chunk-M66SJHQQ.js";
-import "./chunk-EZATO3DW.js";
-import "./chunk-42PJFMAR.js";
+} from "./chunk-KMRFRXIT.js";
+import {
+  DEFAULT_HEADLESS_TIMEOUT_MS,
+  runHeadless
+} from "./chunk-CGKLE6WR.js";
+import "./chunk-LEW4QM74.js";
+import "./chunk-3BWZO4YI.js";
+import "./chunk-OOVG2KBY.js";
+import "./chunk-RVO7L4FX.js";
+import "./chunk-QJ5UZAH5.js";
 import {
   resolveIssuePaths
-} from "./chunk-DIDUVL6O.js";
+} from "./chunk-QK7WUO7A.js";
 import {
   getGlobalTimeout,
   printError,
   printInfo,
   printSuccess
-} from "./chunk-BYZDPR7N.js";
+} from "./chunk-TYRNTSPJ.js";
+import "./chunk-6QYMSNVA.js";
+import "./chunk-3WFZETIM.js";
+import "./chunk-HAKDLR6N.js";
+import "./chunk-BEZ5BA5U.js";
+import "./chunk-67YIU2VB.js";
+import "./chunk-P3LRUNC6.js";
+import "./chunk-G5EK2XRN.js";
 import {
   loadTaskPlan,
   saveTaskPlan
-} from "./chunk-CD5ULRVC.js";
+} from "./chunk-J3SOTUVI.js";
+import "./chunk-LBGHI4S2.js";
+import "./chunk-7Z4Y3VKM.js";
+import "./chunk-XEYH6ZZI.js";
 
 // src/commands/analyze.ts
 import { mkdir, readFile, writeFile } from "fs/promises";
@@ -65,9 +78,11 @@ async function runAnalyze(issue, resolvedIssue) {
     outputFormat: "json",
     allowedTools: ["Bash", "Read", "Glob", "Grep", "Write"],
     addDirs: [paths.issueDir],
-    statusMessage: `Analyzing issue #${issueNumber}...`
+    statusMessage: `Analyzing issue #${issueNumber}...`,
+    phase: "analyze",
+    permission: "read-only"
   });
-  publishPhaseMetrics("analyze", result.cost, startedAtMs);
+  publishPhaseMetrics("analyze", result.cost, startedAtMs, result.agent?.provider);
   if (!result.success) {
     printError(`Analysis failed: ${result.error}`);
     return 1;
@@ -95,4 +110,4 @@ async function runAnalyze(issue, resolvedIssue) {
 export {
   runAnalyze
 };
-//# sourceMappingURL=analyze-OEET7KGE.js.map
+//# sourceMappingURL=analyze-WIND2SEB.js.map

package/dist/analyze-WIND2SEB.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/commands/analyze.ts"],"sourcesContent":["import { mkdir, readFile, writeFile } from 'node:fs/promises';\nimport { DEFAULT_HEADLESS_TIMEOUT_MS, runHeadless } from '../core/headless.js';\nimport { applyPlaceholders, loadPrompt } from '../core/prompt-resolver.js';\nimport { publishPhaseMetrics } from '../core/session-metrics.js';\nimport { loadTaskPlan, saveTaskPlan } from '../core/state-manager.js';\nimport { getGlobalTimeout } from '../core/verbose.js';\nimport { issuePlaceholders, resolveCommandIssue } from '../issues/context.js';\nimport type { ResolvedIssue } from '../issues/types.js';\nimport { resolvePolicyPlaceholders } from '../policy/placeholders.js';\nimport { resolveIssuePaths } from '../storage/resolve.js';\nimport { printError, printInfo, printSuccess } from '../ui/logger.js';\n\nexport async function runAnalyze(issue: string, resolvedIssue?: ResolvedIssue): Promise<number> {\n  const issueNumber = issue.replace(/^#/, '');\n  const paths = await resolveIssuePaths(issueNumber);\n  const analysisPath = paths.analysisFile;\n\n  const resolution = await resolveCommandIssue(issueNumber, resolvedIssue);\n  if (!resolution.ok) {\n    return resolution.code;\n  }\n\n  await mkdir(paths.issueDir, { recursive: true });\n\n  const template = await loadPrompt('analyze');\n  const prompt = applyPlaceholders(template, {\n    // The repository's own conventions. Empty when it declares none, which is\n    // what keeps the rendered prompt identical to the pre-policy one.\n    ...(await resolvePolicyPlaceholders()),\n    __ISSUE_NUMBER__: issueNumber,\n    __ANALYSIS_PATH__: analysisPath,\n    // Last: the Issue content is substituted in but never scanned again, so a\n    // body that happens to contain a placeholder is left untouched.\n    ...issuePlaceholders(resolution.resolved),\n  });\n\n  const startedAtMs = Date.now();\n  const result = await runHeadless({\n    prompt,\n    maxTurns: 30,\n    timeout: getGlobalTimeout() ?? DEFAULT_HEADLESS_TIMEOUT_MS,\n    // json (not text) so the CLI reports usage: the envelope's `result` field\n    // carries the same assistant text this phase already consumed.\n    outputFormat: 'json',\n    allowedTools: ['Bash', 'Read', 'Glob', 'Grep', 'Write'],\n    addDirs: [paths.issueDir],\n    statusMessage: `Analyzing issue #${issueNumber}...`,\n    phase: 'analyze',\n    permission: 'read-only',\n  });\n  // Before the success check: the tokens were spent either way.\n  publishPhaseMetrics('analyze', result.cost, startedAtMs, result.agent?.provider);\n\n  if (!result.success) {\n    printError(`Analysis failed: ${result.error}`);\n    return 1;\n  }\n\n  // Verify the file was created\n  try {\n    const content = await readFile(analysisPath, 'utf-8');\n    if (content.length < 10) {\n      printError('Analysis file was created but appears empty');\n      return 1;\n    }\n  } catch {\n    // File wasn't created by headless — save the result as analysis\n    printInfo('Headless did not create analysis file; saving output directly');\n    await writeFile(analysisPath, result.result, 'utf-8');\n  }\n\n  // Update pipeline state\n  const tasksPath = paths.tasksFile;\n  try {\n    const plan = await loadTaskPlan(tasksPath);\n    plan.pipeline.analyzeCompleted = true;\n    await saveTaskPlan(tasksPath, plan);\n  } catch {\n    // tasks.json may not exist yet — that's OK for standalone analyze\n  }\n\n  printSuccess(`Analysis saved to ${analysisPath}`);\n  return 0;\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAAA,SAAS,OAAO,UAAU,iBAAiB;AAY3C,eAAsB,WAAW,OAAe,eAAgD;AAC9F,QAAM,cAAc,MAAM,QAAQ,MAAM,EAAE;AAC1C,QAAM,QAAQ,MAAM,kBAAkB,WAAW;AACjD,QAAM,eAAe,MAAM;AAE3B,QAAM,aAAa,MAAM,oBAAoB,aAAa,aAAa;AACvE,MAAI,CAAC,WAAW,IAAI;AAClB,WAAO,WAAW;AAAA,EACpB;AAEA,QAAM,MAAM,MAAM,UAAU,EAAE,WAAW,KAAK,CAAC;AAE/C,QAAM,WAAW,MAAM,WAAW,SAAS;AAC3C,QAAM,SAAS,kBAAkB,UAAU;AAAA;AAAA;AAAA,IAGzC,GAAI,MAAM,0BAA0B;AAAA,IACpC,kBAAkB;AAAA,IAClB,mBAAmB;AAAA;AAAA;AAAA,IAGnB,GAAG,kBAAkB,WAAW,QAAQ;AAAA,EAC1C,CAAC;AAED,QAAM,cAAc,KAAK,IAAI;AAC7B,QAAM,SAAS,MAAM,YAAY;AAAA,IAC/B;AAAA,IACA,UAAU;AAAA,IACV,SAAS,iBAAiB,KAAK;AAAA;AAAA;AAAA,IAG/B,cAAc;AAAA,IACd,cAAc,CAAC,QAAQ,QAAQ,QAAQ,QAAQ,OAAO;AAAA,IACtD,SAAS,CAAC,MAAM,QAAQ;AAAA,IACxB,eAAe,oBAAoB,WAAW;AAAA,IAC9C,OAAO;AAAA,IACP,YAAY;AAAA,EACd,CAAC;AAED,sBAAoB,WAAW,OAAO,MAAM,aAAa,OAAO,OAAO,QAAQ;AAE/E,MAAI,CAAC,OAAO,SAAS;AACnB,eAAW,oBAAoB,OAAO,KAAK,EAAE;AAC7C,WAAO;AAAA,EACT;AAGA,MAAI;AACF,UAAM,UAAU,MAAM,SAAS,cAAc,OAAO;AACpD,QAAI,QAAQ,SAAS,IAAI;AACvB,iBAAW,6CAA6C;AACxD,aAAO;AAAA,IACT;AAAA,EACF,QAAQ;AAEN,cAAU,+DAA+D;AACzE,UAAM,UAAU,cAAc,OAAO,QAAQ,OAAO;AAAA,EACtD;AAGA,QAAM,YAAY,MAAM;AACxB,MAAI;AACF,UAAM,OAAO,MAAM,aAAa,SAAS;AACzC,SAAK,SAAS,mBAAmB;AACjC,UAAM,aAAa,WAAW,IAAI;AAAA,EACpC,QAAQ;AAAA,EAER;AAEA,eAAa,qBAAqB,YAAY,EAAE;AAChD,SAAO;AACT;","names":[]}

package/dist/{apply-ZIA3C52S.js → apply-JNOI2LR4.js}

@@ -1,14 +1,25 @@
 #!/usr/bin/env node
 import "./chunk-3D4A4B46.js";
 import {
+  DEFAULT_BRANCH_CONVENTION,
   loadRepositoryPolicy
-} from "./chunk-EZATO3DW.js";
-import "./chunk-42PJFMAR.js";
+} from "./chunk-RVO7L4FX.js";
+import "./chunk-QJ5UZAH5.js";
+import "./chunk-QK7WUO7A.js";
+import "./chunk-TYRNTSPJ.js";
+import "./chunk-6QYMSNVA.js";
+import "./chunk-3WFZETIM.js";
+import "./chunk-HAKDLR6N.js";
 import {
   getProjectRoot
-} from "./chunk-DIDUVL6O.js";
-import "./chunk-BYZDPR7N.js";
-import "./chunk-CD5ULRVC.js";
+} from "./chunk-BEZ5BA5U.js";
+import "./chunk-67YIU2VB.js";
+import "./chunk-P3LRUNC6.js";
+import "./chunk-G5EK2XRN.js";
+import "./chunk-J3SOTUVI.js";
+import "./chunk-LBGHI4S2.js";
+import "./chunk-7Z4Y3VKM.js";
+import "./chunk-XEYH6ZZI.js";
 
 // src/scaffold/apply.ts
 import { access, mkdir, writeFile } from "fs/promises";
@@ -503,7 +514,7 @@ it: the label becomes a second truth that ages on its own.
 
 ## Branches and commits
 
-- Branch: \`issue/{N}-{slug}\`
+- Branch: \`${DEFAULT_BRANCH_CONVENTION}\`
 - Commit: ${DEFAULT_COMMIT_CONVENTION}
 
 The commit type must match the nature of the change. A bug fix committed as
@@ -787,4 +798,4 @@ export {
   planRepositoryScaffold,
   resolveRepositoryState
 };
-//# sourceMappingURL=apply-ZIA3C52S.js.map
+//# sourceMappingURL=apply-JNOI2LR4.js.map