codehere 2.0.2 → 2.0.3

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
@@ -65,7 +65,7 @@ codehere ✕ npm test · 2 failed · agent sent back · 1 of 3
65
65
  codehere ✓ npm test · 48 passed · re-run here · 6.2 s
66
66
  ```
67
67
 
68
- ✓ marks only a test run Codehere just made on your machine. A teammate's "verified" is plain text: it was true on their machine, and anyone with push access can edit a lesson file. The agent is told whether the files the fix changed have changed since.
68
+ ✓ marks only a test run Codehere just made on your machine. A run whose own summary counts a failed test is a failure, even when the command exits 0. A teammate's "verified" is plain text: it was true on their machine, and anyone with push access can edit a lesson file. A tick written into a rule or a lesson reaches the agent as `[tick]`. The agent is told whether the files the fix changed have changed since.
69
69
 
70
70
  Don't want to commit help cards? Copy the printed text into WhatsApp, Discord or Slack instead.
71
71
 
@@ -126,7 +126,7 @@ Everything is plain markdown, so you can read, edit, review and delete it like a
126
126
  - **It only reacts to this session's work.** Changes you had before the session are ignored. If the agent changed nothing, Codehere does nothing.
127
127
  - **It never traps the agent.** At most 3 "not done yet" messages per session, then it lets go.
128
128
  - **It fails open.** If anything breaks, the hook exits quietly and the agent carries on.
129
- - **Small model, small job.** Gemma 4 (via Ollama) is optional and only writes summaries for briefs and help cards, which you read before using. Turn it on with `--gemma`.
129
+ - **Small model, small job.** Gemma 4 (via Ollama) is optional and only writes summaries for briefs and help cards, which you read before using. Turn it on with `--gemma`. The text goes only to a model on this machine: Codehere refuses an `OLLAMA_HOST` on another machine and Ollama cloud models, and uses the plain text instead.
130
130
  - **Only human-confirmed context at session start.** Rules are written by people. Lessons appear only when a matching error shows up, because auto-generated context files can make agents worse.
131
131
 
132
132
  ## Try the demo (no Claude account needed)
@@ -140,7 +140,7 @@ npm test # end-to-end checks
140
140
  ## Known limits
141
141
 
142
142
  - Claude Code, Codex CLI and Copilot CLI. OpenCode is next. Copilot in VS Code reads the same hook files but has not been checked, and Copilot's cloud agent is not supported. See [Which agents](#which-agents) for what works on each.
143
- - Lessons match by error text. When both the failure and the lesson name a failing test, the names must be the same, so a lesson for a failed `assert.equal` reaches only a failing test with the same name. Two tests with the same name in different files or suites still share lessons. A generic error outside a test can bring up a lesson that doesn't fit, which is why the agent gets it as reference, not as an instruction.
143
+ - Lessons match by error text: at least 0.45 of the words in the two error lines must be shared (words in both, divided by words in either). When both the failure and the lesson name a failing test, the names must be the same, so a lesson for a failed `assert.equal` reaches only a failing test with the same name. Two tests with the same name in different files or suites still share lessons. A generic error outside a test can bring up a lesson that doesn't fit, which is why the agent gets it as reference, not as an instruction.
144
144
  - Redaction catches common keys, tokens and emails. Always read a help card before sharing it.
145
145
  - The "done" check needs a test command. Without one, only the tampering check runs.
146
146
  - Approvals are per folder. Each git worktree or extra clone needs its own `codehere approve`.
@@ -153,6 +153,6 @@ npm test # end-to-end checks
153
153
  - Only the test command needs approval. A teammate who sets `verifyOnStop` to `false` in `.codehere/config.json` turns the finish check off for everyone; `codehere status` shows it as off.
154
154
  - Sharing is through git. There is no server, so teammates see updates after `git pull`.
155
155
  - Prompt checks are plain word lists (English and Indonesian). They can miss a short prompt in another language.
156
- - The weakened-test scan catches skipped tests, deleted test files and removed assertions. It does not yet catch an edited expected value; see `docs/overlap-with-check.md`.
156
+ - The weakened-test scan catches skipped and focused tests (`.skip`, `xit`, `{ skip: true }`, `.only`, `fit`), deleted test files, an assertion commented out with `//` or `#`, assertions that cannot fail (`assert.equal(1, 1)`), at least two more assertion lines removed than added, and a `package.json` test script changed to always pass (`|| true`). It does not catch one removed assertion, assertions wrapped in `/* */`, options spread over several lines, an edited expected value, or a `return` above an assertion; see `docs/overlap-with-check.md`.
157
157
 
158
158
  MIT licensed.
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "codehere",
3
- "version": "2.0.2",
3
+ "version": "2.0.3",
4
4
  "description": "A git-native team brain for vibe coders. Silent fixes for AI coding agents: breaks doom loops, verifies 'done', blocks test tampering, and shares what your team learned.",
5
5
  "repository": {
6
6
  "type": "git",
package/src/brain.js CHANGED
@@ -323,8 +323,8 @@ export function gitHashes(root, files) {
323
323
 
324
324
  // A tick is Codehere's mark for a run it made on this machine (DESIGN.md §1). Text from the
325
325
  // repo or from a teammate never brings its own.
326
- export function noTick(text) {
327
- return String(text).replace(/[✓✔☑✅]/g, '?');
326
+ export function noTick(text, mark = '?') {
327
+ return String(text).replace(/[✓✔☑✅√🗸🗹]/gu, mark);
328
328
  }
329
329
 
330
330
  // A commit id read back from state.json, or null. The state file can arrive through git,
package/src/commands.js CHANGED
@@ -496,8 +496,8 @@ export async function brief(flags) {
496
496
  `Rewrite this into one clear, complete instruction for a coding agent. Keep every requirement and constraint. Plain English, max 120 words.\n\n${text}`,
497
497
  cfg.gemmaModel
498
498
  );
499
- if (summary) text = text.replace('## Goal', `## In one paragraph (written by ${summary.model} on this machine; an opinion, check it)\n${summary.text}\n\n## Goal`);
500
- else console.error('(Gemma not reachable through Ollama; using the plain brief.)');
499
+ if (summary?.text) text = text.replace('## Goal', `## In one paragraph (written by ${summary.model} on this machine; an opinion, check it)\n${summary.text}\n\n## Goal`);
500
+ else console.error(summary?.error || '(Gemma not reachable through Ollama; using the plain brief.)');
501
501
  }
502
502
 
503
503
  const file = path.join(paths(root).briefs, `${stamp()}.md`);
@@ -546,8 +546,8 @@ export async function help(question, flags) {
546
546
  `Summarise this help request for a busy teammate in 2 sentences: what is wrong and what you need from them.\n\n${text}`,
547
547
  cfg.gemmaModel
548
548
  );
549
- if (summary) text = `## In short (written by ${summary.model} on this machine; an opinion, check it)\n${summary.text}\n\n${text}`;
550
- else console.error('(Gemma not reachable through Ollama; using the plain help card.)');
549
+ if (summary?.text) text = `## In short (written by ${summary.model} on this machine; an opinion, check it)\n${summary.text}\n\n${text}`;
550
+ else console.error(summary?.error || '(Gemma not reachable through Ollama; using the plain help card.)');
551
551
  }
552
552
 
553
553
  const id = `${stamp()}-${slug(ask, 30)}`;
package/src/gemma.js CHANGED
@@ -4,7 +4,7 @@
4
4
 
5
5
  const url = () => {
6
6
  const host = process.env.OLLAMA_HOST || '127.0.0.1:11434';
7
- return /^https?:\/\//.test(host) ? host : `http://${host}`; // Ollama allows OLLAMA_HOST without a scheme
7
+ return /^https?:\/\//i.test(host) ? host : `http://${host}`; // Ollama allows OLLAMA_HOST without a scheme
8
8
  };
9
9
 
10
10
  async function post(path, body, ms) {
@@ -36,10 +36,24 @@ export function pickModel(installed, wanted) {
36
36
  return null;
37
37
  }
38
38
 
39
- // Returns { text, model } or null.
39
+ // Text goes only to a model on this machine. Ollama forwards a cloud model (`gpt-oss:120b-cloud`,
40
+ // `glm-4.6:cloud`) to its own servers, and OLLAMA_HOST can name another machine.
41
+ const LOCAL_HOST = /^(localhost\.?|127(\.\d{1,3}){3}|0\.0\.0\.0|\[::1?\])$/i;
42
+ const isCloud = (name) => /[:-]cloud$/i.test(name || '');
43
+
44
+ // Returns { text, model }, { error } when Codehere refuses to send the text, or null.
40
45
  export async function summarize(prompt, wanted = 'gemma4:e4b') {
46
+ let host = '';
47
+ try {
48
+ host = new URL(url()).hostname;
49
+ } catch {}
50
+ if (!LOCAL_HOST.test(host)) return { error: `(OLLAMA_HOST is ${process.env.OLLAMA_HOST}, which is not a loopback address. Codehere sends text only to a local model; using the plain text.)` };
51
+ if (isCloud(wanted)) return { error: `(${wanted} is an Ollama cloud model. Codehere sends text only to a local model; using the plain text.)` };
41
52
  const tags = await post('/api/tags', null, 3000);
42
- const model = tags ? pickModel((tags.models || []).map((m) => m.name), wanted) : null;
53
+ const all = (tags && tags.models) || [];
54
+ const local = all.filter((m) => !isCloud(m.name) && !m.remote_host);
55
+ const model = pickModel(local.map((m) => m.name), wanted);
56
+ if (!model && local.length < all.length) return { error: '(No local Gemma model, and Ollama cloud models are not used. Codehere sends text only to a local model; using the plain text.)' };
43
57
  if (!model) return null;
44
58
  // Thinking off: a summary does not need it, and it is several times slower.
45
59
  const data = await post('/api/generate', { model, prompt, stream: false, think: false, options: { temperature: 0.2 } }, 60000);
package/src/hooks.js CHANGED
@@ -145,9 +145,11 @@ function sessionStart(root, input, agent) {
145
145
  }
146
146
 
147
147
  writeState(root, state);
148
+ // Rules, briefs and help cards come from the repo: none of it may show the agent Codehere's tick.
149
+ const text = noTick(out.join('\n'), '[tick]'); // "[tick]" keeps the meaning of a spec that asks for one
148
150
  // Copilot reads only JSON here; Claude Code and Codex take the plain text
149
- if (out.length && agent === 'copilot') emitJson({ additionalContext: out.join('\n') });
150
- if (out.length) process.stdout.write(out.join('\n') + '\n');
151
+ if (out.length && agent === 'copilot') emitJson({ additionalContext: text });
152
+ if (out.length) process.stdout.write(text + '\n');
151
153
  quiet();
152
154
  }
153
155
 
@@ -320,7 +322,7 @@ function lessonText(root, doc) {
320
322
  : 'Proof none recorded.'
321
323
  );
322
324
  out.push('Check whether it fits before trying something new.');
323
- return out.join('\n');
325
+ return noTick(out.join('\n'), '[tick]'); // a teammate's text cannot show the agent Codehere's tick
324
326
  }
325
327
 
326
328
  function lessonLine(doc) {
@@ -549,15 +551,17 @@ function runTests(root, cfg) {
549
551
  });
550
552
  const output = `${r.stdout || ''}\n${r.stderr || ''}`.trim();
551
553
  const timedOut = r.error && r.error.code === 'ETIMEDOUT';
552
- return { ok: r.status === 0 && !timedOut, output, timedOut, code: r.status, seconds: (Date.now() - started) / 1000 };
554
+ // The runner's own count of failures wins over exit 0: `|| true` on a test script is one edit.
555
+ const c = testCounts(output);
556
+ return { ok: r.status === 0 && !timedOut && !(c && c.failed), output, timedOut, code: r.status, seconds: (Date.now() - started) / 1000 };
553
557
  }
554
558
 
555
559
  // "48 passed" or "2 failed" when the runner's summary can be read, else the exit code
556
560
  export function runResult(t) {
557
561
  if (t.timedOut) return 'timed out';
558
562
  const c = testCounts(t.output);
559
- if (t.ok) return c && !c.failed ? `${c.passed} passed` : 'exit 0'; // a summary with failures but exit 0: say only what is sure
560
- return c && c.failed ? `${c.failed} failed` : `exit ${t.code}`;
563
+ if (t.ok) return c && !c.failed ? `${c.passed} passed` : 'exit 0'; // no summary to read: say only the exit code
564
+ return c && c.failed ? `${c.failed} failed${t.code === 0 ? ' · exit 0' : ''}` : `exit ${t.code}`;
561
565
  }
562
566
 
563
567
  const removedLine = (at) => `codehere ✕ codehere uninstall ran at ${new Date(at).toTimeString().slice(0, 5)} · this repo's checks are removed`;
@@ -665,6 +669,7 @@ function stop(root, input) {
665
669
  block(
666
670
  [
667
671
  `Codehere ran \`${cmd}\` and it ${t.timedOut ? 'timed out' : 'failed'}. The task is not done yet.`,
672
+ ...(t.code === 0 && !t.timedOut ? ['Its own summary counts failed tests, though the command exited 0.'] : []),
668
673
  'Last lines of output:',
669
674
  tail,
670
675
  'Fix the cause, then finish. Do not change or skip tests to make them pass.',
package/src/signals.js CHANGED
@@ -133,6 +133,25 @@ const TAMPER_ADDED = [
133
133
  { re: /expect\([^)]*\)\.(toBe|toEqual)\(\s*(true|1)\s*\)\s*;?\s*\/\/\s*(todo|temp|hack)/i, why: 'an assertion was replaced with a placeholder' },
134
134
  ];
135
135
 
136
+ // Only in test files: elsewhere `{ skip: true }` or `assert.equal(1, 1)` can be ordinary code.
137
+ const LITERAL = String.raw`(\d+(?:\.\d+)?|true|false|null|'[^']*'|"[^"]*")`;
138
+ const TAMPER_ADDED_IN_TESTS = [
139
+ // test('name', { skip: true }, ...): the name cannot run past its closing quote, so the options are this test's
140
+ { re: /(?<![.\w])(it|test|describe|suite)\s*\(\s*(['"`])(?:(?!\2)[^\\]|\\.)*\2\s*,\s*\{[^}]*\b(skip|todo)\s*:\s*(true\b|['"`])/, why: 'a test was skipped ({ skip: true })' },
141
+ // fit/fdescribe only in jest's form, starting a statement with a quoted name then a callback:
142
+ // `fit(` alone is a common function name
143
+ { re: /(?<![.\w])(it|test|describe|context|suite)\.only\s*\(|^[+-]\s*(fit|fdescribe)\s*\(\s*(['"`])(?:(?!\3)[^\\]|\\.)*\3\s*,\s*(async\s+)?(\(|function\b|\w+\s*=>)/, why: 'other tests were turned off (.only)' },
144
+ { re: /\bassert(\.ok)?\(\s*true\s*\)|^\+\s*assert\s+True\s*$/, why: 'an assertion that cannot fail was added' },
145
+ { re: new RegExp(String.raw`\bassert\.\w+\(\s*${LITERAL}\s*,\s*\1\s*\)`), why: 'an assertion that cannot fail was added' },
146
+ { re: new RegExp(String.raw`\bexpect\(\s*${LITERAL}\s*\)\.(toBe|toEqual|toStrictEqual)\(\s*\1\s*\)`), why: 'an assertion that cannot fail was added' },
147
+ ];
148
+ // assert(...), assert.equal(...), expect(...), self.assertEqual(...), and pytest's bare `assert x`;
149
+ // not expect.objectContaining(...) and the other matchers inside one assertion
150
+ const ASSERT = /\b(expect\s*\(|assert[\w.]*\s*\()|^[+-]\s*assert\s/;
151
+ // package.json's test script made to pass whatever the tests do: `|| true`, `|| :`, `|| echo`, `; exit 0`
152
+ const ALWAYS_PASSES = /"test"\s*:\s*"[^"]*(\|\|\s*(true|:|exit\s+0|echo\b[^"&|;]*)|;\s*(true|exit\s+0))\s*"/;
153
+ const COMMENT = /^[+-]\s*(\/\/|#|\/\*|\*)\s*/;
154
+
136
155
  export function tamperFindings(nameStatus, diff) {
137
156
  const found = [];
138
157
  for (const line of nameStatus.split('\n')) {
@@ -142,17 +161,32 @@ export function tamperFindings(nameStatus, diff) {
142
161
  let current = '';
143
162
  let removedAsserts = 0;
144
163
  let addedAsserts = 0;
164
+ let removed = new Set(); // this file's removed assertions, to spot one coming back as a comment
165
+ let hadFallback = false; // the test script already ended in `|| true` before this session
145
166
  for (const line of diff.split('\n')) {
146
167
  if (line.startsWith('+++ ')) {
147
168
  current = line.replace(/^\+\+\+ (b\/)?/, '');
169
+ removed = new Set();
170
+ hadFallback = false;
148
171
  continue;
149
172
  }
150
173
  if (line.startsWith('--- ')) continue;
174
+ const inTest = TEST_FILE.test(current);
175
+ const probe = line.slice(0, 10000); // the patterns can take quadratic time on a long line
176
+ if (/(^|\/)package\.json$/.test(current) && ALWAYS_PASSES.test(probe)) {
177
+ if (line.startsWith('-')) hadFallback = true;
178
+ else if (!hadFallback) found.push(`the test script was changed to always exit 0 in ${current}`);
179
+ }
151
180
  if (line.startsWith('+')) {
152
- for (const t of TAMPER_ADDED) if (t.re.test(line)) found.push(`${t.why} in ${current}`);
153
- if (TEST_FILE.test(current) && /\b(expect|assert)\w*\s*\(/.test(line)) addedAsserts++;
154
- } else if (line.startsWith('-')) {
155
- if (TEST_FILE.test(current) && /\b(expect|assert)\w*\s*\(/.test(line)) removedAsserts++;
181
+ for (const t of TAMPER_ADDED) if (t.re.test(probe)) found.push(`${t.why} in ${current}`);
182
+ if (!inTest) continue;
183
+ for (const t of TAMPER_ADDED_IN_TESTS) if (t.re.test(probe)) found.push(`${t.why} in ${current}`);
184
+ if (COMMENT.test(line)) {
185
+ if (removed.has(line.replace(COMMENT, '').trim())) found.push(`an assertion was commented out in ${current}`);
186
+ } else if (ASSERT.test(line)) addedAsserts++;
187
+ } else if (line.startsWith('-') && inTest && !COMMENT.test(line) && ASSERT.test(line)) {
188
+ removedAsserts++;
189
+ removed.add(line.slice(1).trim());
156
190
  }
157
191
  }
158
192
  if (removedAsserts - addedAsserts >= 2) {
@@ -192,6 +226,7 @@ const num = (s, re) => {
192
226
  const m = s.match(re);
193
227
  return m ? Number(m[1]) : null;
194
228
  };
229
+ const after = (t, line) => t.slice(t.indexOf(line) + line.length);
195
230
  const counts = (passed, failed) => (passed == null && failed == null ? null : { passed: passed || 0, failed: failed || 0 });
196
231
  const SUMMARIES = [
197
232
  // node --test, spec and TAP reporters: "ℹ tests 3", "ℹ pass 2", "ℹ fail 1"
@@ -203,15 +238,19 @@ const SUMMARIES = [
203
238
  re: /^[=\s]*(?:\d+ \w+(?:, )?)+ in [\d.]+s\b.*$/gm,
204
239
  read: (t, line) => counts(num(line, /(\d+) passed/), (num(line, /(\d+) failed/) || 0) + (num(line, /(\d+) errors?/) || 0)),
205
240
  },
206
- // mocha "46 passing" and "2 failing"
207
- { re: /^\s*\d+ passing\b.*$/gm, read: (t) => counts(num(t, /^\s*(\d+) passing\b/m), num(t, /^\s*(\d+) failing\b/m)) },
208
- // python unittest "Ran 5 tests in 0.001s", then "OK" or "FAILED (failures=2, errors=1)"
241
+ // mocha "46 passing", then "2 pending" and "2 failing" on the next lines (a log line elsewhere is not the count)
242
+ {
243
+ re: /^\s*\d+ passing\b.*$/gm,
244
+ read: (t, line) => counts(num(line, /(\d+) passing/), num(after(t, line).split('\n').slice(1, 3).join('\n'), /^\s*(\d+) failing\b/m)),
245
+ },
246
+ // python unittest "Ran 5 tests in 0.001s", then the status line: "OK" or "FAILED (failures=2, errors=1)"
209
247
  {
210
248
  re: /^Ran \d+ tests? in [\d.]+s$/gm,
211
- read: (t) => {
212
- if (/expected failures|unexpected successes/.test(t)) return null; // neither passed nor failed
213
- const failed = (num(t, /failures=(\d+)/) || 0) + (num(t, /errors=(\d+)/) || 0);
214
- return counts(num(t, /^Ran (\d+) tests?/m) - failed - (num(t, /skipped=(\d+)/) || 0), failed);
249
+ read: (t, line) => {
250
+ const status = (after(t, line).trimStart().match(/^(OK|FAILED)\b.*/) || [''])[0];
251
+ if (!status || /expected failures|unexpected successes/.test(status)) return null; // neither passed nor failed
252
+ const failed = (num(status, /failures=(\d+)/) || 0) + (num(status, /errors=(\d+)/) || 0);
253
+ return counts(num(line, /^Ran (\d+)/) - failed - (num(status, /skipped=(\d+)/) || 0), failed);
215
254
  },
216
255
  },
217
256
  // cargo "test result: ok. 4 passed; 0 failed", one line per test binary