next-issue 0.8.1 → 0.10.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/README.md CHANGED
@@ -21,8 +21,9 @@ For each open issue, oldest first:
21
21
  1. Claim the issue only when it holds the `status:todo` label, holds no
22
22
  `status:in-progress` or `status:in-review` label, and has no assignee. A
23
23
  `status:done`, `status:needs-human` or `status:blocked` label always stops
24
- the claim. An issue with saved state under `.next-issue/` resumes instead,
25
- whatever its labels, unless another person is now the assignee.
24
+ the claim, a resume too. An issue with saved state under `.next-issue/`
25
+ resumes instead, whatever its other labels, unless another person is now the
26
+ assignee.
26
27
  2. Assign the issue to you and add `status:in-progress`.
27
28
  3. Fetch the remote, fast-forward the local copy of the default branch when
28
29
  that is safe, then add a git worktree at `../<repo>-issue-<n>` on the
@@ -39,13 +40,21 @@ For each open issue, oldest first:
39
40
  7. On a red build, give the failed job logs to the fixer agent, then go to step
40
41
  6 again. The budget is `maxCiFixes`.
41
42
  8. Let the reviewer agent judge the diff, at most `diffMaxChars` of it. The
42
- reviewer returns a structured verdict that rates each finding `blocking` or
43
- `minor`. The harness puts the result on the pull request.
44
- 9. No blocking finding means approval: mark the pull request ready for review
45
- and set `status:done`. The merge stays with you. The worktree, the local
46
- branch and the state file go away. The remote branch stays. A blocking
47
- finding always stops the approval, even when the reviewer also says approve.
48
- 10. With a blocking finding, run the fixer agent and go to step 6 again. The
43
+ reviewer reads files and runs commands in the worktree, so it can prove a
44
+ finding with the tests or the type check. The harness then throws away every
45
+ change that the reviewer left, so no such change can reach a later commit.
46
+ The reviewer returns a structured verdict that rates each finding `blocking`
47
+ or `minor`. The harness puts the result on the pull request.
48
+ 9. An open finding is any finding, blocking or minor, while a review round is
49
+ left. In the last round, only a blocking finding stays open: a minor finding
50
+ there goes on the pull request but starts no fix, because a fix in the last
51
+ round gets no review. Thus `maxReviewRounds` of `1` never fixes a minor
52
+ finding.
53
+ 10. No open finding means approval: mark the pull request ready for review and
54
+ set `status:done`. The merge stays with you. The worktree, the local branch
55
+ and the state file go away. The remote branch stays. An open finding always
56
+ stops the approval, even when the reviewer also says approve.
57
+ 11. With an open finding, run the fixer agent and go to step 6 again. The
49
58
  budget is `maxReviewRounds`.
50
59
 
51
60
  ## Stops for a loop
@@ -85,16 +94,20 @@ too, from the first review round. Thus a draft pull request shows work that the
85
94
  harness has not finished, and a `status:needs-human` pull request shows work
86
95
  that waits for a person.
87
96
 
88
- The harness runs all git and `gh` commands itself. The agents only read and
89
- change files. The commit holds all the changes in the worktree, so a setup
90
- command that writes a file outside `.gitignore` puts that file in the commit.
97
+ The harness runs all git and `gh` commands itself. The commit holds all the
98
+ changes in the worktree, so a setup command that writes a file outside
99
+ `.gitignore` puts that file in the commit. The reviewer is the one agent whose
100
+ changes never get that far: the harness discards them before it reads the
101
+ verdict.
91
102
 
92
103
  ## Safety
93
104
 
94
- The implementer and the fixer run with the permission checks off. They start in
95
- the worktree, but nothing holds them there: they can run any command, with your
96
- rights, anywhere on the machine. The prompt holds the title, the body and the
97
- comments of the issue, which come from GitHub.
105
+ Every agent can run a shell command. The implementer and the fixer run with the
106
+ permission checks off; the reviewer runs with a smaller tool set, but its shell
107
+ has no limit on the commands. They all start in the worktree, but nothing holds
108
+ them there: they can run any command, with your rights, anywhere on the machine.
109
+ The prompt holds the title, the body and the comments of the issue, which come
110
+ from GitHub.
98
111
 
99
112
  Thus a person who can write an issue or a comment can try to give an
100
113
  instruction to the agent. Use the harness only on a repository where you trust
@@ -111,6 +124,9 @@ Every run has an id, `<timestamp>`, and writes two files under
111
124
  one report per issue with the outcome, the reason, the pull request number
112
125
  and the budgets used.
113
126
 
127
+ At the start of a run, the harness deletes the files under `.next-issue/runs/`
128
+ that are older than `keepRunDays`. Set the field to `0` to keep them all.
129
+
114
130
  A short text summary of the same data goes to standard output. Use `--json` to
115
131
  get the full summary object there instead, so the harness fits in a pipe. All
116
132
  progress goes to standard error.
@@ -154,8 +170,8 @@ npx -y next-issue@latest init
154
170
 
155
171
  The command writes `next-issue.config.json` in the root of the repository and
156
172
  adds `.next-issue/` to `.gitignore`. It stops when the config file is there
157
- already; `--force` replaces it. The file holds no `setupCommand`, because that
158
- field has no default.
173
+ already; `--force` replaces it. The file holds no `setupCommand`, because the
174
+ harness deduces one from the lock file of the project.
159
175
 
160
176
  A field that is not in the file keeps its default. A file that is not valid
161
177
  JSON, a field with a value of the wrong type and a field name that the harness
@@ -175,6 +191,7 @@ does not know all stop the run, so a typo cannot pass without a word.
175
191
  "commandTimeoutMinutes": 10,
176
192
  "logMaxChars": 20000,
177
193
  "diffMaxChars": 60000,
194
+ "keepRunDays": 7,
178
195
  "draftPullRequest": true,
179
196
  "models": {},
180
197
  "labels": {
@@ -200,10 +217,11 @@ does not know all stop the run, so a typo cannot pass without a word.
200
217
  | `agentTimeoutMinutes` | `30` | The limit for one agent call |
201
218
  | `setupTimeoutMinutes` | `15` | The limit for the setup command |
202
219
  | `commandTimeoutMinutes` | `10` | The limit for one `git` or `gh` command |
203
- | `logMaxChars` | `20000` | The maximum length of the failed job logs |
220
+ | `logMaxChars` | `20000` | The maximum length of the failed job logs, shared between the failed jobs |
204
221
  | `diffMaxChars` | `60000` | The maximum length of the diff that the reviewer reads |
222
+ | `keepRunDays` | `7` | The number of days to keep the run files; `0` keeps them all |
205
223
  | `draftPullRequest` | `true` | Open the pull request as a draft, until the review approves |
206
- | `setupCommand` | none | A shell command to run in a new worktree, before the implementer |
224
+ | `setupCommand` | from the lock file | A shell command to run in a new worktree, before the implementer |
207
225
  | `models` | `{}` | The model per agent role |
208
226
  | `labels` | see above | The names of the labels that the harness reads and sets |
209
227
 
@@ -213,8 +231,30 @@ claims every open issue that no other label and no assignee holds back.
213
231
  The log gives the reason for a skipped issue as `stop-label`, `not-ready`,
214
232
  `in-flight` or `assigned`.
215
233
 
216
- A role without an entry in `models` uses the default model of the SDK. A
217
- `setupCommand` that fails, or that does not finish in `setupTimeoutMinutes`,
234
+ A role without an entry in `models` uses the default model of the SDK.
235
+
236
+ Without a `setupCommand`, the harness looks in the new worktree for a lock file
237
+ and runs the install command that goes with it:
238
+
239
+ | Lock file | Command |
240
+ | --- | --- |
241
+ | `pnpm-lock.yaml` | `pnpm install --frozen-lockfile` |
242
+ | `package-lock.json` | `npm ci` |
243
+ | `yarn.lock` | `yarn install --frozen-lockfile`, or `yarn install --immutable` with a `.yarnrc.yml` |
244
+ | `bun.lock`, `bun.lockb` | `bun install --frozen-lockfile` |
245
+ | `Cargo.lock` | `cargo fetch` |
246
+ | `go.sum` | `go mod download` |
247
+ | `uv.lock` | `uv sync` |
248
+ | `poetry.lock` | `poetry install` |
249
+ | `Gemfile.lock` | `bundle install` |
250
+
251
+ The first match in that order wins. A project with no lock file in the list
252
+ runs no setup, and a `setupCommand` in the config file always wins. Thus the
253
+ agents find a worktree with its dependencies in place, and the reviewer can run
254
+ the tests. Each command holds to the lock file, so a lock file that does not
255
+ match the manifest stops the issue rather than changing the lock file.
256
+
257
+ A `setupCommand` that fails, or that does not finish in `setupTimeoutMinutes`,
218
258
  hands the issue to a person. The wait for the checks has its own limit, so
219
259
  `commandTimeoutMinutes` holds for every other `git` and `gh` command. State per
220
260
  issue goes to `.next-issue/<issue>.json`, so a new run continues where the last
@@ -228,6 +268,8 @@ would hand the issue over again at once. To give the issue another try, take the
228
268
  and the findings back to zero and keeps the branch, the pull request and the
229
269
  step that the last run reached, so the work carries on where it stopped. A
230
270
  commit that you pushed to the pull request in the meantime is fetched first.
271
+ The flag touches only an issue that the run handles, so the state of an issue
272
+ that a label still holds back stays as it was.
231
273
 
232
274
  ## Commands and options
233
275
 
@@ -243,6 +285,9 @@ commit that you pushed to the pull request in the meantime is fetched first.
243
285
  | `--quiet` | Show only the milestones and the summary |
244
286
  | `--help` | Show the option list |
245
287
 
288
+ The exit code is 1 when an issue needs a person or failed, and 0 when it does
289
+ not, so a cron job or a CI step can read the result of the run.
290
+
246
291
  ## Development
247
292
 
248
293
  ```bash
package/dist/agents.mjs CHANGED
@@ -1,7 +1,7 @@
1
1
  import { query } from "@anthropic-ai/claude-agent-sdk";
2
2
  import { emptyUsage } from "./observe.mjs";
3
- const READ_ONLY = {
4
- tools: ["Read", "Grep", "Glob"],
3
+ const REVIEWING = {
4
+ tools: ["Read", "Grep", "Glob", "Bash"],
5
5
  permissionMode: "dontAsk",
6
6
  };
7
7
  const CODING = {
@@ -9,7 +9,7 @@ const CODING = {
9
9
  permissionMode: "bypassPermissions",
10
10
  allowDangerouslySkipPermissions: true,
11
11
  };
12
- export { CODING, READ_ONLY };
12
+ export { CODING, REVIEWING };
13
13
  export async function runAgent(recorder, request) {
14
14
  const log = recorder.scope({ role: request.name });
15
15
  const usage = emptyUsage();
package/dist/cli.mjs CHANGED
@@ -7,9 +7,9 @@ import { setCommandObserver, setDefaultTimeout } from "./exec.mjs";
7
7
  import { detectRepo, ensureIgnored, repoRoot } from "./git.mjs";
8
8
  import { currentLogin, defaultBranch, openIssues } from "./github.mjs";
9
9
  import { exitCode, runIssues } from "./loop.mjs";
10
- import { message, Recorder } from "./observe.mjs";
10
+ import { message, pruneRuns, Recorder } from "./observe.mjs";
11
11
  import { PORTS, processIssue } from "./pipeline.mjs";
12
- import { resetState, STATE_DIR, takeStop } from "./state.mjs";
12
+ import { STATE_DIR, takeStop } from "./state.mjs";
13
13
  import { formatSummary } from "./summary.mjs";
14
14
  function write(stream, text) {
15
15
  return new Promise((resolve, reject) => {
@@ -30,6 +30,10 @@ async function main() {
30
30
  setDefaultTimeout(config.commandTimeoutMinutes * 60_000);
31
31
  const repo = await detectRepo(root, config.remote);
32
32
  const recorder = await Recorder.create(repo.root, args.level);
33
+ const pruned = await pruneRuns(repo.root, config.keepRunDays);
34
+ if (pruned > 0) {
35
+ recorder.event("runs.pruned", { files: pruned, keepDays: config.keepRunDays }, "verbose");
36
+ }
33
37
  setCommandObserver((record) => {
34
38
  recorder.event(record.code === 0 ? "command" : "command.fail", {
35
39
  cmd: `${record.command} ${record.args.join(" ")}`,
@@ -56,27 +60,26 @@ async function init(root, force) {
56
60
  }
57
61
  async function handle(args, config, repo, recorder) {
58
62
  const [base, login] = await Promise.all([defaultBranch(repo), currentLogin(repo)]);
59
- const context = { repo, config, base, login, recorder, ports: PORTS };
63
+ const context = { repo, config, base, login, reset: args.reset, recorder, ports: PORTS };
60
64
  recorder.event("run.start", { repo: `${repo.owner}/${repo.name}`, base, login, log: recorder.logFile }, "quiet");
61
65
  const issues = await openIssues(repo, config.issueLimit);
62
66
  recorder.event("issues", { open: issues.length }, "quiet");
63
67
  const { reports, stoppedAt } = await runIssues(args, {
64
68
  issues,
65
69
  process: (issue) => processIssue(context, issue),
66
- reset: (issue) => resetState(repo, issue),
67
70
  stop: () => takeStop(repo),
68
71
  });
69
72
  if (stoppedAt !== undefined) {
70
73
  recorder.event("run.stop", { issue: stoppedAt }, "quiet");
71
74
  }
72
- const summary = { ...recorder.summary(), issues: reports };
73
75
  recorder.event("run.end", {
74
76
  done: count(reports, "done"),
75
77
  needsHuman: count(reports, "needs-human"),
76
78
  failed: count(reports, "error"),
77
79
  skipped: count(reports, "skipped"),
78
- tokens: summary.total.total,
80
+ tokens: recorder.summary().total.total,
79
81
  }, "quiet");
82
+ const summary = { ...recorder.summary(), issues: reports };
80
83
  const summaryFile = join(repo.root, STATE_DIR, "runs", `${recorder.runId}.summary.json`);
81
84
  await writeFile(summaryFile, `${JSON.stringify(summary, null, 2)}\n`);
82
85
  await write(process.stdout, args.json ? `${JSON.stringify(summary, null, 2)}\n` : formatSummary(summary));
package/dist/config.mjs CHANGED
@@ -14,6 +14,7 @@ const DEFAULTS = {
14
14
  commandTimeoutMinutes: 10,
15
15
  logMaxChars: 20000,
16
16
  diffMaxChars: 60000,
17
+ keepRunDays: 7,
17
18
  draftPullRequest: true,
18
19
  models: {},
19
20
  labels: {
@@ -38,6 +39,7 @@ const WHOLE_FIELDS = [
38
39
  ["commandTimeoutMinutes", 1],
39
40
  ["logMaxChars", 1],
40
41
  ["diffMaxChars", 1],
42
+ ["keepRunDays", 0],
41
43
  ];
42
44
  const MODEL_ROLES = ["implementer", "reviewer", "fixer"];
43
45
  const LABEL_NAMES = ["ready", "inProgress", "inReview", "done", "needsHuman", "skip"];
package/dist/exec.mjs CHANGED
@@ -1,6 +1,7 @@
1
1
  import { spawn } from "node:child_process";
2
2
  let observer;
3
3
  let defaultTimeoutMs = 10 * 60_000;
4
+ const STDERR_TAIL = 200_000;
4
5
  export function setCommandObserver(next) {
5
6
  observer = next;
6
7
  }
@@ -30,20 +31,19 @@ export function run(command, args, options = {}) {
30
31
  stdout = stdout.slice(-options.tailChars);
31
32
  }
32
33
  });
34
+ const stderrLimit = options.tailChars ?? STDERR_TAIL;
33
35
  child.stderr?.on("data", (chunk) => {
34
36
  stderr += chunk;
37
+ if (stderr.length > stderrLimit) {
38
+ stderr = stderr.slice(-stderrLimit);
39
+ }
35
40
  });
36
41
  const timer = setTimeout(() => {
37
42
  timedOut = true;
38
43
  child.kill("SIGKILL");
39
44
  }, limit);
40
45
  let done = false;
41
- child.on("error", (error) => {
42
- clearTimeout(timer);
43
- done = true;
44
- reject(error);
45
- });
46
- child.on("close", (code) => {
46
+ const settle = (code) => {
47
47
  clearTimeout(timer);
48
48
  if (done) {
49
49
  return;
@@ -60,7 +60,21 @@ export function run(command, args, options = {}) {
60
60
  timedOut,
61
61
  });
62
62
  resolve(result);
63
+ };
64
+ child.on("error", (error) => {
65
+ clearTimeout(timer);
66
+ done = true;
67
+ reject(error);
68
+ });
69
+ child.on("exit", (code) => {
70
+ if (!timedOut) {
71
+ return;
72
+ }
73
+ child.stdout?.destroy();
74
+ child.stderr?.destroy();
75
+ settle(code);
63
76
  });
77
+ child.on("close", settle);
64
78
  if (options.input !== undefined) {
65
79
  child.stdin?.on("error", () => undefined);
66
80
  child.stdin?.end(options.input);
package/dist/git.mjs CHANGED
@@ -100,6 +100,14 @@ export async function commitAll(cwd, message) {
100
100
  await must("git", ["commit", "-m", message], { cwd });
101
101
  return true;
102
102
  }
103
+ export async function discardChanges(cwd) {
104
+ if (!(await isDirty(cwd))) {
105
+ return false;
106
+ }
107
+ await must("git", ["checkout", "--", "."], { cwd });
108
+ await must("git", ["clean", "-fd"], { cwd });
109
+ return true;
110
+ }
103
111
  export async function push(cwd, remote, branch) {
104
112
  await must("git", ["push", "-u", remote, branch], { cwd });
105
113
  }
package/dist/github.mjs CHANGED
@@ -78,8 +78,12 @@ export async function ensureLabel(repo, label) {
78
78
  if (known.has(label)) {
79
79
  return;
80
80
  }
81
- await run("gh", ["label", "create", label, "--repo", slug(repo)], { cwd: repo.root });
82
- known.add(label);
81
+ const result = await run("gh", ["label", "create", label, "--repo", slug(repo)], {
82
+ cwd: repo.root,
83
+ });
84
+ if (result.code === 0) {
85
+ known.add(label);
86
+ }
83
87
  }
84
88
  export async function findPr(repo, branch) {
85
89
  const result = await run("gh", ["pr", "list", "--head", branch, "--state", "open", "--json", "number", "--repo", slug(repo)], { cwd: repo.root });
@@ -90,7 +94,7 @@ export async function findPr(repo, branch) {
90
94
  return parsed[0]?.number;
91
95
  }
92
96
  export async function createPr(repo, branch, base, title, body, draft) {
93
- await gh(repo, [
97
+ const created = await gh(repo, [
94
98
  "pr",
95
99
  "create",
96
100
  "--head",
@@ -103,12 +107,16 @@ export async function createPr(repo, branch, base, title, body, draft) {
103
107
  body,
104
108
  ...(draft ? ["--draft"] : []),
105
109
  ]);
106
- const number = await findPr(repo, branch);
110
+ const number = prNumber(created) ?? (await findPr(repo, branch));
107
111
  if (number === undefined) {
108
112
  throw new Error(`The pull request for ${branch} was not found after creation.`);
109
113
  }
110
114
  return number;
111
115
  }
116
+ export function prNumber(output) {
117
+ const match = /\/pull\/(\d+)\b/.exec(output);
118
+ return match === null ? undefined : Number(match[1]);
119
+ }
112
120
  async function prHead(repo, branch) {
113
121
  return gh(repo, ["pr", "view", branch, "--json", "headRefOid", "--jq", ".headRefOid"]);
114
122
  }
@@ -178,11 +186,11 @@ async function read(probe, head, timeoutMs, shown) {
178
186
  if ((await probe.list()) === "none") {
179
187
  return "none";
180
188
  }
189
+ return await probe.watch(timeoutMs);
181
190
  }
182
191
  catch (error) {
183
192
  return error instanceof Error ? error : new Error(String(error));
184
193
  }
185
- return probe.watch(timeoutMs);
186
194
  }
187
195
  export function waitForChecks(repo, branch, options) {
188
196
  return pollChecks({
@@ -197,6 +205,7 @@ export async function failedCheckLogs(repo, branch, maxChars) {
197
205
  return "The checks failed, but gh reports no check for the branch.";
198
206
  }
199
207
  const failed = checks.filter((check) => check.bucket === "fail");
208
+ const share = Math.max(1, Math.floor(maxChars / Math.max(1, failed.length)));
200
209
  const parts = [];
201
210
  for (const check of failed) {
202
211
  parts.push(`## ${check.name}\n${check.description}`);
@@ -206,11 +215,11 @@ export async function failedCheckLogs(repo, branch, maxChars) {
206
215
  }
207
216
  const logs = await run("gh", ["run", "view", runId, "--log-failed", "--repo", slug(repo)], {
208
217
  cwd: repo.root,
209
- tailChars: maxChars,
218
+ tailChars: share,
210
219
  });
211
- parts.push(logs.stdout.slice(-maxChars));
220
+ parts.push(logs.stdout.slice(-share));
212
221
  }
213
- return parts.join("\n\n").slice(-maxChars);
222
+ return parts.join("\n\n");
214
223
  }
215
224
  export async function markPrReady(repo, pr) {
216
225
  const result = await run("gh", ["pr", "ready", String(pr), "--repo", slug(repo)], {
package/dist/loop.mjs CHANGED
@@ -5,9 +5,6 @@ export async function runIssues(args, loop) {
5
5
  if (args.max !== undefined && handled(reports) >= args.max) {
6
6
  break;
7
7
  }
8
- if (args.reset) {
9
- await loop.reset(issue.number);
10
- }
11
8
  const report = await loop.process(issue);
12
9
  reports.push(report);
13
10
  if (await loop.stop()) {
package/dist/observe.mjs CHANGED
@@ -1,5 +1,5 @@
1
1
  import { createWriteStream } from "node:fs";
2
- import { mkdir } from "node:fs/promises";
2
+ import { mkdir, readdir, rm, stat } from "node:fs/promises";
3
3
  import { join } from "node:path";
4
4
  import { STATE_DIR } from "./state.mjs";
5
5
  const VISIBILITY = { quiet: 0, normal: 1, verbose: 2 };
@@ -117,6 +117,35 @@ export class Recorder {
117
117
  await this.sink.close();
118
118
  }
119
119
  }
120
+ export async function pruneRuns(root, keepDays) {
121
+ if (keepDays <= 0) {
122
+ return 0;
123
+ }
124
+ const dir = join(root, STATE_DIR, "runs");
125
+ let names;
126
+ try {
127
+ names = await readdir(dir);
128
+ }
129
+ catch {
130
+ return 0;
131
+ }
132
+ const cutoff = Date.now() - keepDays * 24 * 60 * 60_000;
133
+ let dropped = 0;
134
+ for (const name of names) {
135
+ const path = join(dir, name);
136
+ try {
137
+ const info = await stat(path);
138
+ if (info.isFile() && info.mtimeMs < cutoff) {
139
+ await rm(path, { force: true });
140
+ dropped += 1;
141
+ }
142
+ }
143
+ catch {
144
+ continue;
145
+ }
146
+ }
147
+ return dropped;
148
+ }
120
149
  export function message(error) {
121
150
  return error instanceof Error ? error.message : String(error);
122
151
  }
package/dist/pipeline.mjs CHANGED
@@ -1,12 +1,13 @@
1
- import { CODING, READ_ONLY, runAgent } from "./agents.mjs";
1
+ import { CODING, REVIEWING, runAgent } from "./agents.mjs";
2
2
  import { managedLabels } from "./config.mjs";
3
3
  import { run } from "./exec.mjs";
4
- import { addWorktree, branchName, commitAll, deleteBranch, diff, hasWorktree, push, removeWorktree, revision, worktreePath, } from "./git.mjs";
4
+ import { addWorktree, branchName, commitAll, deleteBranch, diff, discardChanges, hasWorktree, push, removeWorktree, revision, worktreePath, } from "./git.mjs";
5
5
  import { assignIssue, commentOnPr, createPr, failedCheckLogs, findPr, issueComments, markPrReady, setLabel, waitForChecks, } from "./github.mjs";
6
6
  import { message } from "./observe.mjs";
7
7
  import { capDiff, fixPrompt, implementPrompt, parseCommitSubject, parseUnrelated, reviewPrompt, } from "./prompts.mjs";
8
- import { dropState, readState, writeState } from "./state.mjs";
9
- import { blockingFindings, fingerprint, formatFindings, formatVerdict, isApproved, readVerdict, VERDICT_SCHEMA, } from "./verdict.mjs";
8
+ import { detectSetupCommand } from "./setup.mjs";
9
+ import { dropState, readState, resetState, writeState, } from "./state.mjs";
10
+ import { blockingFindings, fingerprint, formatFindings, formatVerdict, openFindings, readVerdict, VERDICT_SCHEMA, } from "./verdict.mjs";
10
11
  export const PORTS = {
11
12
  addWorktree,
12
13
  assignIssue,
@@ -14,7 +15,9 @@ export const PORTS = {
14
15
  commitAll,
15
16
  createPr,
16
17
  deleteBranch,
18
+ detectSetupCommand,
17
19
  diff,
20
+ discardChanges,
18
21
  failedCheckLogs,
19
22
  findPr,
20
23
  hasWorktree,
@@ -55,7 +58,7 @@ export async function processIssue(context, issue) {
55
58
  return { issue: issue.number, outcome: "skipped", ciFixes: 0, reviewRounds: 0, ms: 0 };
56
59
  }
57
60
  log.event("issue.start", { title: issue.title, createdAt: issue.createdAt }, "quiet");
58
- const state = saved ?? {
61
+ const fresh = {
59
62
  issue: issue.number,
60
63
  phase: "claimed",
61
64
  branch: branchName(issue.number),
@@ -63,6 +66,7 @@ export async function processIssue(context, issue) {
63
66
  reviewRounds: 0,
64
67
  reviewLog: [],
65
68
  };
69
+ const state = saved === undefined ? fresh : context.reset ? resetState(saved) : saved;
66
70
  try {
67
71
  const report = await work(context, issue, log, state);
68
72
  log.event("issue.end", { outcome: report.outcome, reason: report.reason, ms: report.ms }, "quiet");
@@ -119,8 +123,9 @@ async function work(context, issue, log, state) {
119
123
  ms: Date.now() - started,
120
124
  };
121
125
  };
122
- if (config.setupCommand !== undefined) {
123
- const setup = await log.step("setup", { cmd: config.setupCommand }, () => ports.runSetup(config.setupCommand, worktree, config.setupTimeoutMinutes * 60_000));
126
+ const setupCommand = config.setupCommand ?? (await ports.detectSetupCommand(worktree));
127
+ if (setupCommand !== undefined) {
128
+ const setup = await log.step("setup", { cmd: setupCommand, found: config.setupCommand === undefined }, () => ports.runSetup(setupCommand, worktree, config.setupTimeoutMinutes * 60_000));
124
129
  if (setup.code !== 0) {
125
130
  const late = setup.timedOut === true;
126
131
  log.event("setup.fail", { code: setup.code, timedOut: late, stderr: setup.stderr.trim().slice(-2000) }, "quiet");
@@ -223,26 +228,31 @@ async function work(context, issue, log, state) {
223
228
  name: "reviewer",
224
229
  cwd: worktree,
225
230
  prompt: reviewPrompt(issue, comments, patch, history(state.reviewLog)),
226
- profile: READ_ONLY,
231
+ profile: REVIEWING,
227
232
  model: config.models.reviewer,
228
233
  timeoutMs: config.agentTimeoutMinutes * 60_000,
229
234
  outputSchema: VERDICT_SCHEMA,
230
235
  }));
236
+ if (await ports.discardChanges(worktree)) {
237
+ log.event("review.discard", { round: state.reviewRounds }, "quiet");
238
+ }
231
239
  const verdict = readVerdict(review.structured);
232
240
  if (verdict === undefined) {
233
241
  await escalate("The reviewer gave no verdict.");
234
242
  return finish("needs-human", "no verdict");
235
243
  }
236
244
  const blocking = blockingFindings(verdict);
245
+ const open = openFindings(verdict, state.reviewRounds >= config.maxReviewRounds);
237
246
  log.event("verdict", {
238
247
  round: state.reviewRounds,
239
248
  verdict: verdict.verdict,
240
249
  blocking: blocking.length,
241
250
  minor: verdict.findings.length - blocking.length,
251
+ open: open.length,
242
252
  summary: verdict.summary,
243
253
  }, "quiet");
244
- await ports.commentOnPr(repo, state.pr, formatVerdict(verdict));
245
- if (isApproved(verdict)) {
254
+ await ports.commentOnPr(repo, state.pr, formatVerdict(verdict, open));
255
+ if (open.length === 0) {
246
256
  const after = await gate();
247
257
  if (after.done) {
248
258
  return after.report;
@@ -256,15 +266,15 @@ async function work(context, issue, log, state) {
256
266
  await setStatus(context, issue.number, state.pr, config.labels.done);
257
267
  return finish("done");
258
268
  }
259
- const mark = fingerprint(blocking);
269
+ const mark = fingerprint(open);
260
270
  if (state.reviewLog.some((round) => round.fingerprint === mark)) {
261
271
  await escalate("The reviewer repeated findings that an earlier round did not fix.");
262
272
  return finish("needs-human", "repeated findings");
263
273
  }
264
274
  const earlier = history(state.reviewLog);
265
- state.reviewLog.push({ round: state.reviewRounds, fingerprint: mark, findings: formatFindings(blocking) });
275
+ state.reviewLog.push({ round: state.reviewRounds, fingerprint: mark, findings: formatFindings(open) });
266
276
  await writeState(repo, state);
267
- const fix = await fixRound(job, "The reviewer requested changes.", formatFindings(blocking), earlier, "review");
277
+ const fix = await fixRound(job, "The reviewer requested changes.", formatFindings(open), earlier, "review");
268
278
  if (!fix.committed) {
269
279
  await escalate("The fixer added no commit for the review findings.");
270
280
  return finish("needs-human", "no fix commit");
package/dist/prompts.mjs CHANGED
@@ -43,7 +43,9 @@ export function reviewPrompt(issue, comments, diff, earlier) {
43
43
  diff,
44
44
  "```",
45
45
  "## Rules",
46
- "Read files for context when you must. Do not change any file.",
46
+ "Read files and run commands to prove what you say: the tests, the type check or the linter.",
47
+ "The worktree is ready to use. Do not change a file. Do not write with git or gh.",
48
+ "The harness discards each change that you make, so a change is only lost work.",
47
49
  scope,
48
50
  "Mark a finding blocking only for a wrong result, a missing part of the issue, a regression or a broken convention.",
49
51
  "Give the result as the structured output: the verdict, a short summary and one entry per finding.",
package/dist/setup.mjs ADDED
@@ -0,0 +1,27 @@
1
+ import { access } from "node:fs/promises";
2
+ import { join } from "node:path";
3
+ const MARKERS = [
4
+ ["pnpm-lock.yaml", "pnpm install --frozen-lockfile"],
5
+ ["package-lock.json", "npm ci"],
6
+ [".yarnrc.yml", "yarn install --immutable"],
7
+ ["yarn.lock", "yarn install --frozen-lockfile"],
8
+ ["bun.lock", "bun install --frozen-lockfile"],
9
+ ["bun.lockb", "bun install --frozen-lockfile"],
10
+ ["Cargo.lock", "cargo fetch"],
11
+ ["go.sum", "go mod download"],
12
+ ["uv.lock", "uv sync"],
13
+ ["poetry.lock", "poetry install"],
14
+ ["Gemfile.lock", "bundle install"],
15
+ ];
16
+ export async function detectSetupCommand(root) {
17
+ for (const [file, command] of MARKERS) {
18
+ try {
19
+ await access(join(root, file));
20
+ return command;
21
+ }
22
+ catch {
23
+ continue;
24
+ }
25
+ }
26
+ return undefined;
27
+ }
package/dist/state.mjs CHANGED
@@ -15,15 +15,10 @@ export async function readState(repo, issue) {
15
15
  return undefined;
16
16
  }
17
17
  }
18
- export async function resetState(repo, issue) {
19
- const state = await readState(repo, issue);
20
- if (state === undefined) {
21
- return false;
22
- }
18
+ export function resetState(state) {
23
19
  const next = { ...state, ciFixes: 0, reviewRounds: 0, reviewLog: [] };
24
20
  delete next.handedOver;
25
- await writeState(repo, next);
26
- return true;
21
+ return next;
27
22
  }
28
23
  export async function writeState(repo, state) {
29
24
  await mkdir(dir(repo), { recursive: true });
package/dist/verdict.mjs CHANGED
@@ -55,13 +55,18 @@ export function readVerdict(value) {
55
55
  export function blockingFindings(verdict) {
56
56
  return verdict.findings.filter((finding) => finding.severity === "blocking");
57
57
  }
58
- export function isApproved(verdict) {
59
- return blockingFindings(verdict).length === 0;
58
+ export function openFindings(verdict, lastRound) {
59
+ return lastRound ? blockingFindings(verdict) : verdict.findings;
60
60
  }
61
- export function formatVerdict(verdict) {
62
- const head = isApproved(verdict) ? "Review: approved" : "Review: changes requested";
61
+ const LEFT = "The findings above stay as they are: no blocking finding is open and the review budget is spent.";
62
+ export function formatVerdict(verdict, open) {
63
+ const approved = open.length === 0;
64
+ const head = approved ? "Review: approved" : "Review: changes requested";
63
65
  const lines = verdict.findings.map((finding) => `- **${finding.severity}** ${finding.detail}`);
64
- return [`### ${head}`, verdict.summary, lines.join("\n")].filter((part) => part.length > 0).join("\n\n");
66
+ const note = approved && verdict.findings.length > 0 ? LEFT : "";
67
+ return [`### ${head}`, verdict.summary, lines.join("\n"), note]
68
+ .filter((part) => part.length > 0)
69
+ .join("\n\n");
65
70
  }
66
71
  export function formatFindings(findings) {
67
72
  return findings.map((finding) => `- ${finding.detail}`).join("\n");
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "next-issue",
3
- "version": "0.8.1",
3
+ "version": "0.10.0",
4
4
  "description": "Autonomous issue harness driven by the Claude Agent SDK",
5
5
  "keywords": [
6
6
  "github",
@@ -37,10 +37,10 @@
37
37
  "prepublishOnly": "npm run typecheck && npm test"
38
38
  },
39
39
  "devDependencies": {
40
- "@types/node": "^26.2.0",
40
+ "@types/node": "^26.4.1",
41
41
  "typescript": "^7.0.2"
42
42
  },
43
43
  "dependencies": {
44
- "@anthropic-ai/claude-agent-sdk": "^0.3.241"
44
+ "@anthropic-ai/claude-agent-sdk": "^0.3.261"
45
45
  }
46
46
  }