# Show HN: Aisync exits 3 when two AI agents claim the same file > Source: > Published: 2026-11-13 14:00:05+00:00 ## Repository navigation ## Expand file tree / Copy path # collab.py More file actions 476 lines (414 loc) · 24.6 KB / Copy path # collab.py ## File metadata and controls 476 lines (414 loc) · 24.6 KB 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 """aisync collaboration layer: shared rules any agent can update, claims, progress fan-out, report. What: the pieces that make the hub the one core every agent follows. - rule_add: any agent (Claude Code, Codex, Cursor, ...) adds a shared rule with one command; it lands in preferences.md (the truth) and the next cycle copies it to every tool. - harvest_rules: new timestamped entries in the "## Meta rule" section of the AGENTS.md of each repo listed in `rule_sources` are imported automatically, so a rule an agent wrote there (and forgot to `aisync rule add`) still reaches every tool. - claims: "I am working on these files of this repo now", with a TTL; overlaps warn. - render_progress: the rules pointer + live progress written into files each tool auto-loads (e.g. a block in ~/.codex/AGENTS.md), so an agent sees it without being told to look. - report: read-only self-check (unattributed commits, silent tools, missing projects, stale blockers, failures). It only reports; it never changes a rule by itself. Why: one shared rule set that any agent can update and every agent obeys. How: every function here is called from core.run_cycle (isolated steps) or from the CLI. """ from __future__ import annotations import re from datetime import datetime, timedelta from pathlib import Path from . import core, ruleids RULES_HEAD = "### Rules added by agents (via aisync, newest last)" META_HEAD = "## Meta rule" _ENTRY = re.compile(r"^- (\d{4}-\d{2}-\d{2} \d{2}:\d{2} [A-Z]{2,5}):\s*(.*)$") CLAIM_TTL_MIN = 120 def _norm(text: str) -> str: return re.sub(r"[^a-z0-9\u4e00-\u9fff]+", "", text.lower()) def _stamp() -> str: return datetime.now().astimezone().strftime("%Y-%m-%d %H:%M %Z") def _tilde(p: Path | str) -> str: return str(p).replace(str(Path.home()), "~", 1) # ───────────────────────────── shared rules ───────────────────────────── def rule_lines(block: str) -> list[str]: return [l for l in block.splitlines() if l.startswith("- ")] def _insert_rule(cfg: core.Config, bullet: str, key_text: str) -> str: """Append one bullet, with the next free ID, inside the owner-preferences block. Returns the bullet as written, or "" if an equal rule exists.""" path = cfg.root / "preferences.md" text = path.read_text() block = core.extract_block(text) if block is None: raise ValueError("preferences.md has no owner-preferences block") key = _norm(key_text)[:60] if key and any(key in _norm(l) for l in rule_lines(block)): return "" bullet = ruleids.with_id(bullet, ruleids.next_id(cfg)) body = block[: -len(core.BLOCK_END)].rstrip("\n") if RULES_HEAD not in body: body += f"\n\n{RULES_HEAD}\n" new_block = body + "\n" + bullet + "\n" + core.BLOCK_END core.atomic_write(path, core.replace_block(text, new_block)) return bullet SUPERSEDED_HEAD = "### 已取代" def topics(block: str) -> list[str]: """Topic headings of a tidied rules block (### headings other than 已取代 and the legacy list).""" return [l[4:].strip() for l in block.splitlines() if l.startswith("### ") and l.strip() != SUPERSEDED_HEAD and l.strip() != RULES_HEAD] def _sentences(text: str) -> int: """Sentence ends are . ! ? followed by a space or the end, and 。！？ anywhere. A dot inside a file name or date (x.env, x_autopilot.py, 2026-10-27) and semicolons do not end a sentence (2026-10-27: three false rejections).""" return len([x for x in re.split(r"(?<=[.!?])(?=\s|$)\s*|(?<=[。！？])\s*", text) if x.strip()]) def _insert_topic_rule(cfg: core.Config, topic: str, bullet: str, key_text: str, replaces: str, new_name: str) -> dict: """Put a rule at the top of its topic with the next free ID; the rule it replaces (given by ID like R-087, or by a piece of its text) moves to rules-archive.md with its retired date and the new rule's ID.""" path = cfg.root / "preferences.md" text = path.read_text() block = core.extract_block(text) if block is None: raise ValueError("preferences.md has no owner-preferences block") names = topics(block) hit = [t for t in names if topic.lower() in t.lower()] if len(hit) != 1: raise ValueError(f"--topic must match exactly one topic; topics: {' | '.join(names)}") lines = block.splitlines() span = None if replaces: rid = ruleids.is_id(replaces) cur = [i for i, l in enumerate(lines) if l.startswith("- ") and not _in_superseded(lines, i) and (ruleids.bullet_id(l) == rid if rid else replaces in l)] if len(cur) != 1: what = f"rule {rid}" if rid else "current rule" raise ValueError(f"--replaces must match exactly one {what} (matched {len(cur)})") i = cur[0] end = i + 1 while end < len(lines) and lines[end].startswith(" ") and lines[end].strip(): end += 1 span = (i, end) # Duplicate check: only current rules count, minus the one being replaced. A revision usually # keeps the old rule's opening words, and 已取代 lines are history, not rules. key = _norm(key_text)[:60] live = [l for k, l in enumerate(lines) if l.startswith("- ") and not _in_superseded(lines, k) and not (span and span[0] <= k < span[1])] if key and any(key in _norm(l) for l in live): return {"added": False, "moved": None} new_id = ruleids.next_id(cfg) bullet = ruleids.with_id(bullet, new_id) moved = None if span: i, end = span old = " ".join(x.strip() for x in lines[i:end]) if not ruleids.bullet_id(old): # a rule written before IDs existed gets one on the way out old = ruleids.with_id(old, ruleids.next_id(cfg)) day = f"{datetime.now().astimezone():%Y-%m-%d}" moved = ruleids.archive_line(old, day, new_id, note=f" 被 {new_id}「{new_name}」取代（{day}）") del lines[i:end] h = lines.index(f"### {hit[0]}") lines.insert(h + 1, bullet) if moved: ruleids.archive_add(cfg, [moved]) # history first: a crash in between never loses the old text core.atomic_write(path, core.replace_block(text, "\n".join(lines))) return {"added": True, "moved": moved, "id": new_id, "bullet": bullet} def _in_superseded(lines: list[str], i: int) -> bool: for l in reversed(lines[:i]): if l.startswith("### "): return l.strip() == SUPERSEDED_HEAD return False def rule_add(cfg: core.Config, *, agent: str, text: str, reason: str = "", project: str = "", topic: str = "", replaces: str = "", enforced_by: str = "", judgment_only: str = "", planned: str = "", size: str = "") -> dict: """Add a shared rule. Once preferences.md is grouped into topics, the writing standard applies: pick a topic, write only the current rule in at most three sentences, end with the source tag, and move the rule it replaces (by ID or text) into rules-archive.md in the same write. Every new rule gets the next free ID. The reason goes to the journal only. Before the tidy (no topic headings) it appends to the legacy list as before. A new rule may name what enforces it (enforced_by paths, a judgment_only reason, or the planned check that is not written yet; one of them is required with `require_enforcer = true`); it is written to rules-enforcement.toml under the new ID, never into the rules block. """ from . import enforcement agent, text = (agent or "").strip(), " ".join((text or "").split()) if not agent: raise ValueError("agent is required (pass --agent or set AISYNC_AGENT)") if len(text) < 8: raise ValueError("rule text is too short") entry = (enforcement.check_args(enforced_by, judgment_only, planned, size) if cfg.require_enforcer or enforced_by or judgment_only or planned else None) # optional unless require_enforcer old_id = ruleids.is_id(replaces) if replaces else None old_plan = enforcement.load(cfg).get(old_id, {}) if old_id else {} if old_plan.get("planned") and (entry or {}).get("judgment_only"): # a replacement must not drop a planned check raise ValueError(f"{old_id} has a planned check ({str(old_plan['planned'])[:80]}); the rule that replaces it " "keeps it: pass --planned \"\" --size S|M|L, or --enforced-by ") if (old_plan.get("planned") and (entry or {}).get("planned") # nor postpone it: a larger size moves the due day and enforcement.SIZES[entry["size"]] > enforcement.SIZES.get(str(old_plan.get("size", "")), 0)): raise ValueError(f"{old_id} is planned {old_plan.get('size', '?')} since {old_plan.get('since', '?')}; a rule that " f"replaces it keeps that size (pass --size {old_plan.get('size', 'S')}). To move the due day " "afterwards: aisync rule enforce --planned \"\" --size --reason \"\"") who = agent + (f", {project}" if project else "") block = core.extract_block((cfg.root / "preferences.md").read_text()) or "" if topics(block): if not topic: raise ValueError(f"--topic is required; topics: {' | '.join(topics(block))}") if _sentences(text) > 3 and not cfg.rule_spec: raise ValueError("a rule is at most three sentences; put the why in --reason") if cfg.rule_spec: from . import rulespec bad = rulespec.problems(text, need_id=False, need_tag=False) if bad: raise ValueError("rule is not an interface (" + "; ".join(bad) + "); write " "'When: . Do: . Not: .' or pass --when/--do/--not") bullet = f"- {text.rstrip()} ({datetime.now().astimezone():%Y-%m-%d} · {who})" with core.locked(cfg): rep = _insert_topic_rule(cfg, topic, bullet, text, replaces, text[:30]) added = rep["added"] bullet = rep.get("bullet", bullet) else: bullet = f"- {_stamp()} ({who}): {text}" + (f" Reason: {' '.join(reason.split())}" if reason.strip() else "") with core.locked(cfg): written = _insert_rule(cfg, bullet, text) added, bullet, rep = bool(written), written or bullet, {"moved": None} if added and entry is not None: with core.locked(cfg): moved_id = ruleids.bullet_id(rep["moved"]) if rep.get("moved") else None was = enforcement.load(cfg).get(moved_id, {}) if moved_id else {} if was.get("planned") and entry.get("planned"): # the clock of a replaced rule keeps running entry["since"] = min(str(entry["since"]), str(was.get("since") or entry["since"])) if enforcement.SIZES[entry["size"]] > enforcement.SIZES.get(str(was.get("size", "")), 1 << 30): entry["size"] = was["size"] # replaced by a piece of its text: the size cannot grow here elif was.get("planned") and entry.get("judgment_only"): # replaced by a piece of its text: leave a trace entry["note"] = (f"replaces {moved_id}, which had a planned check " f"({was.get('size', '?')}, since {was.get('since', '?')}): {was['planned']}") enforcement.put(cfg, ruleids.bullet_id(bullet), entry) if added: why = f" Reason: {' '.join(reason.split())}" if reason.strip() and topics(block) else "" core.append_event(cfg, agent=agent, kind="rule_change", project=project, summary=f"shared rule added: [{ruleids.bullet_id(bullet)}] {text}{why}" + (f" (replaces [{ruleids.bullet_id(rep['moved'])}])" if rep.get("moved") else ""), files=["preferences.md"] + ([ruleids.ARCHIVE] if rep.get("moved") else [])) return {"added": added, "rule": bullet, "moved": rep.get("moved"), "id": ruleids.bullet_id(bullet) if added else None} def rule_list(cfg: core.Config, archived: bool = False) -> list[str]: """Current rules (each starts with its ID), or the retired ones from rules-archive.md.""" if archived: return ruleids.archive_entries(cfg) block = core.extract_block((cfg.root / "preferences.md").read_text()) or "" return rule_lines(block) # ───────────────────────────── rules written in repo AGENTS.md ───────────────────────────── def rule_source_files(cfg: core.Config) -> list[Path]: """AGENTS.md of every checkout/worktree of each rule-source repo (main checkout + *.wt-*).""" out = [] for proj in core.project_list(cfg): for repo in cfg.rule_sources: if proj.name == repo or proj.name.startswith(repo + "."): if (proj / "AGENTS.md").exists(): out.append(proj / "AGENTS.md") return out def parse_meta_entries(text: str) -> list[tuple[str, str]]: """(timestamp, full text) for each `- YYYY-MM-DD HH:MM TZ: ...` bullet under '## Meta rule'.""" entries, inside, cur = [], False, None for line in text.splitlines(): if line.startswith("## "): inside = line.startswith(META_HEAD) if cur: entries.append(cur) cur = None continue if not inside: continue m = _ENTRY.match(line) if m: if cur: entries.append(cur) cur = [m.group(1), m.group(2)] elif cur and line.startswith(" ") and line.strip(): cur[1] += " " + line.strip() elif cur and not line.strip(): entries.append(cur) cur = None if cur: entries.append(cur) return [(ts, " ".join(t.split())) for ts, t in entries] def _first_sentence(text: str, n: int = 220) -> str: m = re.match(r"(.+?[.;。])(\s|$)", text) s = m.group(1) if m else text return s if len(s) <= n else s[: n - 1] + "…" def harvest_rules(cfg: core.Config) -> dict: """Import NEW meta-rule entries. A repo seen for the first time only records what exists.""" st = core.load_json(cfg.state / "rule_sources.json", {"repos": [], "seen": []}) seen, repos = set(st["seen"]), set(st["repos"]) rep = {"files": 0, "imported": [], "errors": []} first_sight = set() for path in rule_source_files(cfg): repo = path.parent.name.split(".")[0] try: entries = parse_meta_entries(path.read_text()) except OSError as ex: rep["errors"].append(f"{path.parent.name}: {type(ex).__name__}") continue rep["files"] += 1 if repo not in repos: first_sight.add(repo) for ts, text in entries: key = core._h(ts + _norm(text)[:40]) if key in seen: continue seen.add(key) if repo in repos: # an entry that appeared after we started watching summary = _first_sentence(text) main = path.parent.parent / repo # worktrees come and go; point at the main checkout src = main / "AGENTS.md" if (main / "AGENTS.md").exists() else path bullet = f"- {ts} (from {repo} AGENTS.md): {summary} Full text: `{_tilde(src)}`." with core.locked(cfg): added = _insert_rule(cfg, bullet, summary) if added: # imports are never blocked; they count as none until someone says from . import enforcement enforcement.put(cfg, ruleids.bullet_id(added), {"none": f"imported from {repo} AGENTS.md without an enforcer"}) if added: core.append_event(cfg, agent=f"repo:{repo}", kind="rule_change", project=repo, summary=f"shared rule imported from {repo} AGENTS.md: {summary}", files=["preferences.md"]) rep["imported"].append(summary) repos |= first_sight core.save_json(cfg.state / "rule_sources.json", {"repos": sorted(repos), "seen": sorted(seen)}) return rep # ───────────────────────────── claims (who is working where, now) ───────────────────────────── def _now() -> datetime: return datetime.now().astimezone() def load_claims(cfg: core.Config) -> list[dict]: now = _now() out = [] for c in core.load_json(cfg.state / "claims.json", []): try: if datetime.fromisoformat(c["expires"]) > now: out.append(c) except (KeyError, ValueError): continue return out def _overlaps(a: dict, b: dict) -> bool: if a["project"] != b["project"] or a["agent"] == b["agent"]: return False fa, fb = set(a.get("files") or []), set(b.get("files") or []) if not fa or not fb: # a claim on the whole repo overlaps everything return True return bool(fa & fb) or any(x.rstrip("/") + "/" in y or y.rstrip("/") + "/" in x for x in fa for y in fb) def find_overlaps(cfg: core.Config, agent: str, project: str, files) -> list[dict]: probe = {"agent": agent, "project": project, "files": list(files or [])} return [c for c in load_claims(cfg) if _overlaps(probe, c)] def claim(cfg: core.Config, *, agent: str, project: str, summary: str, files=(), ttl_min: int = CLAIM_TTL_MIN) -> dict: if not (agent and project and summary): raise ValueError("claim needs --agent, --project and --summary") files = [str(f) for f in files] with core.locked(cfg): claims = [c for c in load_claims(cfg) if not (c["agent"] == agent and c["project"] == project)] new = {"agent": agent, "project": project, "files": files, "summary": summary.strip(), "since": core.now_iso(), "expires": (_now() + timedelta(minutes=ttl_min)).isoformat(timespec="seconds")} overlaps = [c for c in claims if _overlaps(new, c)] claims.append(new) core.save_json(cfg.state / "claims.json", claims) core.append_event(cfg, agent=agent, kind="claim", project=project, summary=f"claimed: {summary}", files=files, refs={"overlap_with": ",".join(sorted({c['agent'] for c in overlaps}))} if overlaps else {}) return {"claim": new, "overlaps": overlaps} def release(cfg: core.Config, *, agent: str, project: str = "") -> list[dict]: with core.locked(cfg): claims = load_claims(cfg) gone = [c for c in claims if c["agent"] == agent and (not project or c["project"] == project)] core.save_json(cfg.state / "claims.json", [c for c in claims if c not in gone]) for c in gone: core.append_event(cfg, agent=agent, kind="release", project=c["project"], summary=f"released: {c['summary']}") return gone # ───────────────────────────── self-check report (read-only) ───────────────────────────── def report(cfg: core.Config, days: int = 7) -> dict: """Counts + plain hints. Never acts on them: too little data makes auto-tuning learn noise.""" since = _now() - timedelta(days=days) events = [] for e in core.read_events(cfg): try: if datetime.fromisoformat(e["ts"]) >= since: events.append(e) except (KeyError, ValueError): continue per_agent, unattributed, no_project, commits_by = {}, {}, 0, {} for e in events: a = e.get("agent", "?") per_agent.setdefault(a, {}).setdefault(e.get("kind", "?"), 0) per_agent[a][e.get("kind", "?")] += 1 if e.get("kind") == "commit": commits_by[a] = commits_by.get(a, 0) + 1 if a.startswith("git:"): unattributed[a[4:]] = unattributed.get(a[4:], 0) + 1 elif not e.get("project") and not a.startswith("repo:"): no_project += 1 runs = [r for r in core.read_runs(cfg, 500) if "steps" in r] failed = sum(1 for r in runs if not r.get("ok")) cur = core.build_current(core.read_events(cfg)) stale = [b for b in cur["open_blocked"] if b.get("ts") and datetime.fromisoformat(b["ts"]) < _now() - timedelta(hours=24)] hints = [] for author, n in sorted(unattributed.items(), key=lambda kv: -kv[1]): hints.append(f"{n} commit(s) by git author '{author}' name no tool: add an `Agent: ` trailer to commits") for tool in cfg.roster: if tool not in per_agent: hints.append(f"{tool} logged nothing in {days} days: it may not be reading {_tilde(cfg.root / 'agents.md')}") elif per_agent[tool].get("commit") and not any(k for k in per_agent[tool] if k not in ("commit", "claim", "release")): hints.append(f"{tool} committed but never ran `aisync log`") if no_project: hints.append(f"{no_project} event(s) without --project") if stale: hints.append(f"{len(stale)} blocker(s) open for more than 24 h") if failed: hints.append(f"{failed} of {len(runs)} sync cycles failed: see `aisync status`") conf = core.load_json(cfg.state / "conflicts.json", {}) if conf: hints.append(f"rule conflict open in {', '.join(conf)}: `aisync resolve`") return {"days": days, "events": len(events), "per_agent": per_agent, "unattributed_commits": unattributed, "events_without_project": no_project, "stale_blockers": len(stale), "cycles": len(runs), "failed_cycles": failed, "open_conflicts": sorted(conf), "rules": len(rule_list(cfg)), "hints": hints} # ───────────────────────────── progress fan-out to files tools auto-load ───────────────────────────── def progress_text(cfg: core.Config, cur: dict, claims: list[dict], hints: list[str]) -> str: lines = [core.PROG_START, f"## Live progress of all AI tools (aisync, auto-generated {core.now_iso()}, do not edit)", "", f"Shared rules: `{_tilde(cfg.root / 'preferences.md')}` (every agent follows them; change one with " f"`aisync rule add --agent \"...\"`). Full hub page: `{_tilde(cfg.root / 'agents.md')}`.", ""] if claims: lines.append("**Working right now** (do not edit these files without coordinating)") lines += [f"- {c['agent']} @ {c['project']}: {core._short(c['summary'], 90)}" f"{' [' + ', '.join(c['files'][:6]) + ']' if c['files'] else ' [whole repo]'} until {c['expires'][11:16]}" for c in claims] lines.append("") lines.append("**Last seen per agent**") for a, e in sorted(cur["agents"].items(), key=lambda kv: kv[1]["ts"] or "", reverse=True)[:8]: lines.append(f"- {a} {e['ts']} · {e['project'] or '-'}: {core._short(e['summary'], 100)}") lines += ["", "**Latest**"] + [f"- {e['ts']} · {e['agent']} · {e['kind']} · {e['project'] or '-'}: {core._short(e['summary'], 100)}" for e in reversed(cur["tail"][-6:])] if hints: lines += ["", "**Hints from the self-check** (`aisync report`)"] + [f"- {h}" for h in hints[:4]] lines.append(core.PROG_END) return "\n".join(lines) def _body(section: str) -> list[str]: """Section minus its timestamped header line, for change detection.""" return [l for l in section.splitlines() if "auto-generated" not in l] def render_progress(cfg: core.Config) -> dict: cur = core.build_current(core.read_events(cfg)) section = progress_text(cfg, cur, load_claims(cfg), report(cfg)["hints"]) rep = {"updated": [], "errors": []} for t in cfg.progress_targets: try: if t.mode == "file": old = t.path.read_text() if t.path.exists() else "" if _body(old) != _body(section + "\n"): core.atomic_write(t.path, section + "\n") rep["updated"].append(t.label) continue text = t.path.read_text() s, e = text.find(core.PROG_START), text.find(core.PROG_END) if s < 0 or e < 0: new = text.rstrip("\n") + "\n\n" + section + "\n" else: if _body(text[s:e + len(core.PROG_END)]) == _body(section): continue new = text[:s] + section + text[e + len(core.PROG_END):] core.atomic_write(t.path, new) rep["updated"].append(t.label) except OSError as ex: rep["errors"].append(f"{t.label}: {type(ex).__name__}") # one bad target never blocks the rest return rep