simplicio-cli 0.4.1__tar.gz → 0.4.3__tar.gz

{simplicio_cli-0.4.1/simplicio_cli.egg-info → simplicio_cli-0.4.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: simplicio-cli
-Version: 0.4.1
+Version: 0.4.3
 Summary: Portable task-to-code pipeline that works with any LLM. Turn a one-line task into a verified code change — diff + test + verify loop. +55 pts on a 156-check benchmark, 21% faster, ~same tokens.
 Author-email: Wesley Simplicio <wesleybob4@gmail.com>
 License: MIT
@@ -31,8 +31,8 @@ Requires-Dist: sentence-transformers>=2.2
 Requires-Dist: numpy>=1.23
 Requires-Dist: anthropic>=0.30
 Requires-Dist: openai>=1.30
-Requires-Dist: simplicio-mapper>=0.6.0
-Requires-Dist: simplicio-prompt>=1.9.0
+Requires-Dist: simplicio-mapper>=0.6.1
+Requires-Dist: simplicio-prompt>=1.12.0
 Requires-Dist: httpx>=0.27
 Requires-Dist: orjson>=3.10
 Requires-Dist: diskcache>=5.6
@@ -134,12 +134,25 @@ M1 MacBook (8 GB), five sub-4B tiny models, six frontier 2026 models, and three
 mid-tier 7B–12B open models. Every one gained at least **+14 points** when
 wrapped in simplicio's 6-layer contract.
 
-#### Hugging Face — Qwen2.5-Coder, re-run on 2026-05-27 (latest mapper, 10 cases/side, 156 checks)
+#### Hugging Face — recommended Qwen3-Coder defaults (HF router)
 
-First batch of the smaller→larger re-benchmark against the latest
-`simplicio-mapper` artifacts. The 1.5B runs on CPU via `transformers`
-(Hugging Face Inference Providers does not serve it); the 3B and 7B run
-through the HF router (`https://router.huggingface.co/v1`).
+The served Qwen Coder recommendation now uses the Qwen3-Coder MoE family.
+`Qwen/Qwen2.5-Coder-3B-Instruct` and
+`Qwen/Qwen2.5-Coder-7B-Instruct` remain available as legacy fallback models for
+historical comparisons and hardware that cannot host the MoE successors.
+
+| Slot | Recommended model | Route | Notes |
+|---|---|---|---|
+| Efficient coder | `Qwen/Qwen3-Coder-30B-A3B-Instruct` | HF router | 30B total / ~3B active MoE successor to the 3B slot |
+| High-ceiling coder | `Qwen/Qwen3-Coder-Next` | HF router | 80B total / ~3B active MoE successor to the 7B slot |
+
+> Reproduce the new default set:
+> `BENCH_BASE_URL=https://router.huggingface.co/v1 BENCH_API_KEY=<hf-token>
+> BENCH_MODELS="Qwen/Qwen3-Coder-30B-A3B-Instruct,Qwen/Qwen3-Coder-Next"
+> python3 bench/run_offline.py`.
+
+Legacy Qwen2.5-Coder baseline, re-run on 2026-05-27 against the latest
+`simplicio-mapper` artifacts (10 cases/side, 156 checks):
 
 | Model | Without simplicio | With simplicio | Gain |
 |---|---|---|---|
@@ -148,10 +161,9 @@ through the HF router (`https://router.huggingface.co/v1`).
 | **Qwen 2.5 Coder 1.5B** (`Qwen/Qwen2.5-Coder-1.5B-Instruct`, local CPU) | 30% | **92%** | **+62 pts** |
 | **HF avg (3 models · 10 cases · 156 checks)** | **34%** | **94%** | **+60 pts (+172%)** |
 
-> Monotonic from smaller to larger: pass-rate with simplicio climbs **92% →
-> 94% → 96%** as the model grows, while the raw-prompt baseline stays at
-> **30–38%**. The 1.5B model gains the most (**+62 pts**) — the contract does
-> the heaviest lifting where the model is weakest. Reproduce:
+> Monotonic from smaller to larger in the legacy baseline: pass-rate with
+> simplicio climbs **92% → 94% → 96%** as the model grows, while the raw-prompt
+> baseline stays at **30–38%**. Reproduce the legacy set:
 > `BENCH_BASE_URL=https://router.huggingface.co/v1 BENCH_API_KEY=<hf-token>
 > BENCH_MODELS="local:Qwen/Qwen2.5-Coder-1.5B-Instruct,Qwen/Qwen2.5-Coder-3B-Instruct,Qwen/Qwen2.5-Coder-7B-Instruct"
 > python3 bench/run_offline.py`.
@@ -167,7 +179,18 @@ Pro) show `n/a` for the new column: their OpenRouter calls hit account-level
 HTTP 402 / provider failures on >50% of requests this round, so the sample is
 too small to publish; their old numbers still stand.
 
-#### Local offline — qwen2.5-coder on Ollama, M1 8 GB, run on 2026-05-27 (30 runs/side, 156 checks)
+#### Local offline — Qwen3-Coder GGUF recommendation, Qwen2.5 legacy baseline
+
+For local OpenAI-compatible servers, prefer the Qwen3-Coder GGUF builds when
+the machine can host MoE weights:
+
+| Slot | Recommended local weights | Notes |
+|---|---|---|
+| Efficient coder | `unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF` | Primary local successor for the 3B-active slot |
+| High-ceiling coder | `unsloth/Qwen3-Coder-Next-GGUF` | 24 GB GPU-class successor for long-context work |
+
+The last fully offline fallback baseline remains qwen2.5-coder on Ollama,
+M1 8 GB, run on 2026-05-27 (30 runs/side, 156 checks):
 
 | Model | Without simplicio | With simplicio | Gain |
 |---|---|---|---|
@@ -180,7 +203,7 @@ too small to publish; their old numbers still stand.
 > `http://localhost:11434/v1` (Ollama's OpenAI-compatible endpoint). A
 > 1.5B-param model running on a 4-year-old laptop reaches **88%** pass-rate
 > with simplicio's contract — same hardware, same model, raw prompt = 32%.
-> Reproduce: `BENCH_BASE_URL=http://localhost:11434/v1 BENCH_API_KEY=ollama
+> Reproduce the legacy fallback: `BENCH_BASE_URL=http://localhost:11434/v1 BENCH_API_KEY=ollama
 > BENCH_MODELS="qwen2.5-coder:7b" python3 bench/run_offline.py`.
 
 #### Tiny models — sub-4B, run on 2026-05-26 (50 runs/side, 260 checks)
@@ -382,6 +405,29 @@ simplicio task "..." --stack angular --target ...
 
 How it works: simplicio shells out to `claude -p "<prompt>"` (or `codex exec "<prompt>"`) as a subprocess, captures stdout, runs the test loop. The inner CLI authenticates via your existing OAuth session in `~/.claude/` or `~/.codex/`. simplicio sets `SIMPLICIO_HOOK_GUARD=1` in the subprocess env so the inner Claude Code session does **not** re-fire simplicio's own UserPromptSubmit hook (no infinite recursion).
 
+For orchestrators such as SendSprint, `simplicio task` also has a structured
+contract:
+
+```bash
+simplicio task "hide Delete button for non-admins" \
+  --stack angular \
+  --target src/app/screen/screen.component.html \
+  --dry-run-task \
+  --json
+
+simplicio task "front-only task" \
+  --stack angular \
+  --target src/app/screen/screen.component.html \
+  --bound-paths "src/app/**" \
+  --json
+```
+
+`--dry-run-task` generates the would-be diff/test output without applying or
+testing it. `--json` returns `{task_id, applied, files_changed, tokens_used,
+cost_usd, diff_summary, warnings}`. Repeat `--bound-paths <glob>` to reject
+diffs outside the allowed edit surface; violations are reported in `warnings`
+and the command exits non-zero.
+
 ### Path 3 example — standalone with API key
 
 ```bash

{simplicio_cli-0.4.1 → simplicio_cli-0.4.3}/README.md

@@ -92,12 +92,25 @@ M1 MacBook (8 GB), five sub-4B tiny models, six frontier 2026 models, and three
 mid-tier 7B–12B open models. Every one gained at least **+14 points** when
 wrapped in simplicio's 6-layer contract.
 
-#### Hugging Face — Qwen2.5-Coder, re-run on 2026-05-27 (latest mapper, 10 cases/side, 156 checks)
+#### Hugging Face — recommended Qwen3-Coder defaults (HF router)
 
-First batch of the smaller→larger re-benchmark against the latest
-`simplicio-mapper` artifacts. The 1.5B runs on CPU via `transformers`
-(Hugging Face Inference Providers does not serve it); the 3B and 7B run
-through the HF router (`https://router.huggingface.co/v1`).
+The served Qwen Coder recommendation now uses the Qwen3-Coder MoE family.
+`Qwen/Qwen2.5-Coder-3B-Instruct` and
+`Qwen/Qwen2.5-Coder-7B-Instruct` remain available as legacy fallback models for
+historical comparisons and hardware that cannot host the MoE successors.
+
+| Slot | Recommended model | Route | Notes |
+|---|---|---|---|
+| Efficient coder | `Qwen/Qwen3-Coder-30B-A3B-Instruct` | HF router | 30B total / ~3B active MoE successor to the 3B slot |
+| High-ceiling coder | `Qwen/Qwen3-Coder-Next` | HF router | 80B total / ~3B active MoE successor to the 7B slot |
+
+> Reproduce the new default set:
+> `BENCH_BASE_URL=https://router.huggingface.co/v1 BENCH_API_KEY=<hf-token>
+> BENCH_MODELS="Qwen/Qwen3-Coder-30B-A3B-Instruct,Qwen/Qwen3-Coder-Next"
+> python3 bench/run_offline.py`.
+
+Legacy Qwen2.5-Coder baseline, re-run on 2026-05-27 against the latest
+`simplicio-mapper` artifacts (10 cases/side, 156 checks):
 
 | Model | Without simplicio | With simplicio | Gain |
 |---|---|---|---|
@@ -106,10 +119,9 @@ through the HF router (`https://router.huggingface.co/v1`).
 | **Qwen 2.5 Coder 1.5B** (`Qwen/Qwen2.5-Coder-1.5B-Instruct`, local CPU) | 30% | **92%** | **+62 pts** |
 | **HF avg (3 models · 10 cases · 156 checks)** | **34%** | **94%** | **+60 pts (+172%)** |
 
-> Monotonic from smaller to larger: pass-rate with simplicio climbs **92% →
-> 94% → 96%** as the model grows, while the raw-prompt baseline stays at
-> **30–38%**. The 1.5B model gains the most (**+62 pts**) — the contract does
-> the heaviest lifting where the model is weakest. Reproduce:
+> Monotonic from smaller to larger in the legacy baseline: pass-rate with
+> simplicio climbs **92% → 94% → 96%** as the model grows, while the raw-prompt
+> baseline stays at **30–38%**. Reproduce the legacy set:
 > `BENCH_BASE_URL=https://router.huggingface.co/v1 BENCH_API_KEY=<hf-token>
 > BENCH_MODELS="local:Qwen/Qwen2.5-Coder-1.5B-Instruct,Qwen/Qwen2.5-Coder-3B-Instruct,Qwen/Qwen2.5-Coder-7B-Instruct"
 > python3 bench/run_offline.py`.
@@ -125,7 +137,18 @@ Pro) show `n/a` for the new column: their OpenRouter calls hit account-level
 HTTP 402 / provider failures on >50% of requests this round, so the sample is
 too small to publish; their old numbers still stand.
 
-#### Local offline — qwen2.5-coder on Ollama, M1 8 GB, run on 2026-05-27 (30 runs/side, 156 checks)
+#### Local offline — Qwen3-Coder GGUF recommendation, Qwen2.5 legacy baseline
+
+For local OpenAI-compatible servers, prefer the Qwen3-Coder GGUF builds when
+the machine can host MoE weights:
+
+| Slot | Recommended local weights | Notes |
+|---|---|---|
+| Efficient coder | `unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF` | Primary local successor for the 3B-active slot |
+| High-ceiling coder | `unsloth/Qwen3-Coder-Next-GGUF` | 24 GB GPU-class successor for long-context work |
+
+The last fully offline fallback baseline remains qwen2.5-coder on Ollama,
+M1 8 GB, run on 2026-05-27 (30 runs/side, 156 checks):
 
 | Model | Without simplicio | With simplicio | Gain |
 |---|---|---|---|
@@ -138,7 +161,7 @@ too small to publish; their old numbers still stand.
 > `http://localhost:11434/v1` (Ollama's OpenAI-compatible endpoint). A
 > 1.5B-param model running on a 4-year-old laptop reaches **88%** pass-rate
 > with simplicio's contract — same hardware, same model, raw prompt = 32%.
-> Reproduce: `BENCH_BASE_URL=http://localhost:11434/v1 BENCH_API_KEY=ollama
+> Reproduce the legacy fallback: `BENCH_BASE_URL=http://localhost:11434/v1 BENCH_API_KEY=ollama
 > BENCH_MODELS="qwen2.5-coder:7b" python3 bench/run_offline.py`.
 
 #### Tiny models — sub-4B, run on 2026-05-26 (50 runs/side, 260 checks)
@@ -340,6 +363,29 @@ simplicio task "..." --stack angular --target ...
 
 How it works: simplicio shells out to `claude -p "<prompt>"` (or `codex exec "<prompt>"`) as a subprocess, captures stdout, runs the test loop. The inner CLI authenticates via your existing OAuth session in `~/.claude/` or `~/.codex/`. simplicio sets `SIMPLICIO_HOOK_GUARD=1` in the subprocess env so the inner Claude Code session does **not** re-fire simplicio's own UserPromptSubmit hook (no infinite recursion).
 
+For orchestrators such as SendSprint, `simplicio task` also has a structured
+contract:
+
+```bash
+simplicio task "hide Delete button for non-admins" \
+  --stack angular \
+  --target src/app/screen/screen.component.html \
+  --dry-run-task \
+  --json
+
+simplicio task "front-only task" \
+  --stack angular \
+  --target src/app/screen/screen.component.html \
+  --bound-paths "src/app/**" \
+  --json
+```
+
+`--dry-run-task` generates the would-be diff/test output without applying or
+testing it. `--json` returns `{task_id, applied, files_changed, tokens_used,
+cost_usd, diff_summary, warnings}`. Repeat `--bound-paths <glob>` to reject
+diffs outside the allowed edit surface; violations are reported in `warnings`
+and the command exits non-zero.
+
 ### Path 3 example — standalone with API key
 
 ```bash

{simplicio_cli-0.4.1 → simplicio_cli-0.4.3}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "simplicio-cli"
-version = "0.4.1"
+version = "0.4.3"
 description = "Portable task-to-code pipeline that works with any LLM. Turn a one-line task into a verified code change — diff + test + verify loop. +55 pts on a 156-check benchmark, 21% faster, ~same tokens."
 readme = "README.md"
 license = { text = "MIT" }
@@ -45,8 +45,8 @@ dependencies = [
   "numpy>=1.23",
   "anthropic>=0.30",
   "openai>=1.30",
-  "simplicio-mapper>=0.6.0",
-  "simplicio-prompt>=1.9.0",
+  "simplicio-mapper>=0.6.1",
+  "simplicio-prompt>=1.12.0",
   "httpx>=0.27",
   "orjson>=3.10",
   "diskcache>=5.6",

simplicio_cli-0.4.3/simplicio/__init__.py

@@ -0,0 +1 @@
+__version__ = "0.4.3"

{simplicio_cli-0.4.1 → simplicio_cli-0.4.3}/simplicio/cli.py

@@ -12,6 +12,7 @@ first CLI use instead — the closest equivalent that works on every machine.
 from __future__ import annotations
 
 import argparse
+import json
 import os
 import sys
 from pathlib import Path
@@ -27,7 +28,8 @@ def maybe_autoinstall(cmd: str | None) -> bool:
         return False
     if cmd in ("init", "detect"):
         return False
-    claude_home = Path.home() / ".claude"
+    home = Path(os.environ["HOME"]) if os.environ.get("HOME") else Path.home()
+    claude_home = home / ".claude"
     if not claude_home.is_dir():
         return False
     hook_path = claude_home / "hooks" / "simplicio-userpromptsubmit.sh"
@@ -50,7 +52,7 @@ def maybe_autoinstall(cmd: str | None) -> bool:
     return False
 
 
-def main():
+def main(argv=None):
     ap = argparse.ArgumentParser(prog="simplicio")
     sub = ap.add_subparsers(dest="cmd", required=True)
 
@@ -63,6 +65,12 @@ def main():
     pt.add_argument("--target", required=True)
     pt.add_argument("--criteria", default="- true state\n- false state")
     pt.add_argument("--constraints", default="- build passes")
+    pt.add_argument("--dry-run-task", action="store_true",
+                    help="generate the would-be task output without applying/testing")
+    pt.add_argument("--json", action="store_true",
+                    help="emit stable structured task output")
+    pt.add_argument("--bound-paths", action="append", default=[],
+                    help="glob limiting which paths the task may change; repeatable")
 
 
     pb = sub.add_parser("bench", help="compare with vs without (real numbers)")
@@ -81,7 +89,7 @@ def main():
     p_det.add_argument("--quiet", action="store_true")
     p_det.add_argument("--json", action="store_true")
 
-    a = ap.parse_args()
+    a = ap.parse_args(argv)
     maybe_autoinstall(a.cmd)
     if a.cmd == "index":
         from .precedent import index_repo
@@ -113,8 +121,30 @@ def main():
             argv += ["--json"]
         return detect_main(argv)
     else:
-        from .pipeline import run
-        run(a.root, a.stack, a.goal, a.target, a.criteria, a.constraints)
+        from .pipeline import run, run_task
+        if a.json or a.dry_run_task:
+            result = run_task(
+                a.root,
+                a.stack,
+                a.goal,
+                a.target,
+                a.criteria,
+                a.constraints,
+                dry_run_task=a.dry_run_task,
+                bound_paths=a.bound_paths,
+                quiet=a.json,
+            )
+            if a.json:
+                print(json.dumps(result, sort_keys=True))
+            else:
+                status = "DRY-RUN" if a.dry_run_task else "DONE"
+                print(f"{status}: {result['diff_summary']}")
+                for warning in result["warnings"]:
+                    print(f"warning: {warning}", file=sys.stderr)
+            return 0 if (a.dry_run_task or result["applied"]) else 1
+        run(a.root, a.stack, a.goal, a.target, a.criteria, a.constraints,
+            bound_paths=a.bound_paths)
+        return 0
 
 if __name__ == "__main__":
     main()

{simplicio_cli-0.4.1 → simplicio_cli-0.4.3}/simplicio/pipeline.py

@@ -1,5 +1,6 @@
 """pipeline.py — build -> generate -> validate -> test -> fix (loop)."""
 from dataclasses import dataclass
+import fnmatch
 import os, re, subprocess
 from .observability import estimate_tokens, log_run
 from .prompt import build_prompt
@@ -18,7 +19,40 @@ class FailureClassification:
     kind: str
     guidance: str
 
-def validate_generated_output(output):
+def extract_changed_files(output):
+    text = output or ""
+    files = []
+    for match in re.finditer(r"^diff --git a/(.+?) b/(.+?)$", text, flags=re.M):
+        files.append(match.group(2).strip())
+    for match in re.finditer(r"^\+\+\+ b/(.+?)$", text, flags=re.M):
+        files.append(match.group(1).strip())
+    return list(dict.fromkeys(f for f in files if f and f != "/dev/null"))
+
+def _matches_bound(path, patterns):
+    normalized = path.replace(os.sep, "/").lstrip("./")
+    for raw in patterns or []:
+        pattern = str(raw).replace(os.sep, "/").lstrip("./")
+        if fnmatch.fnmatch(normalized, pattern):
+            return True
+        if pattern.endswith("/**"):
+            prefix = pattern[:-3].rstrip("/")
+            if normalized == prefix or normalized.startswith(f"{prefix}/"):
+                return True
+    return False
+
+def _bound_path_warnings(files, bound_paths):
+    if not bound_paths:
+        return []
+    outside = [path for path in files if not _matches_bound(path, bound_paths)]
+    if not outside:
+        return []
+    return [
+        "diff touches path outside bound paths: "
+        + ", ".join(outside)
+        + f" (allowed: {', '.join(bound_paths)})"
+    ]
+
+def validate_generated_output(output, bound_paths=None):
     text = output or ""
     hints = []
     has_diff = bool(re.search(r"^diff --git |^--- .+\n\+\+\+ ", text, flags=re.M))
@@ -29,6 +63,7 @@ def validate_generated_output(output):
         hints.append("include a TEST block or concrete test code")
     if re.search(r"(?i)\b(pseudocode|placeholder|todo: implement)\b", text):
         hints.append("replace placeholders with executable code")
+    hints.extend(_bound_path_warnings(extract_changed_files(output), bound_paths))
     return ValidationResult(
         ok=not hints,
         reason="ok" if not hints else "; ".join(hints),
@@ -64,10 +99,10 @@ def build_retry_feedback(attempt, validation=None, test_log=""):
     lines.append("Return the full corrected DIFF + TEST block only.")
     return "\n".join(lines)
 
-def _apply_and_test(output, root):
+def _apply_and_test(output, root, bound_paths=None):
     os.makedirs(os.path.join(root, ".simplicio"), exist_ok=True)
     open(os.path.join(root, ".simplicio/last_output.txt"), "w").write(output or "")
-    validation = validate_generated_output(output)
+    validation = validate_generated_output(output, bound_paths)
     if not validation.ok:
         return False, f"pre-apply validation failed: {validation.reason}"
     # PLUG: extract diff -> git apply; extract test. Here we run the test command.
@@ -75,13 +110,47 @@ def _apply_and_test(output, root):
     p = subprocess.run(cmd, shell=True, cwd=root, capture_output=True, text=True)
     return p.returncode == 0, (p.stdout + p.stderr)[-2000:]
 
-def run(root, stack, goal, target, criteria, constraints):
+def _diff_summary(files_changed):
+    if not files_changed:
+        return "no changed files reported"
+    return "changed " + ", ".join(files_changed)
+
+def _task_result(task_id, prompt, output, *, applied, warnings=None):
+    files_changed = extract_changed_files(output)
+    return {
+        "task_id": task_id,
+        "applied": bool(applied),
+        "files_changed": files_changed,
+        "tokens_used": {
+            "prompt": estimate_tokens(prompt),
+            "completion": estimate_tokens(output or ""),
+        },
+        "cost_usd": 0.0,
+        "diff_summary": _diff_summary(files_changed),
+        "warnings": warnings or [],
+    }
+
+def run_task(root, stack, goal, target, criteria, constraints, *,
+             dry_run_task=False, bound_paths=None, quiet=False):
     prompt = build_prompt(root, stack, goal, target, criteria, constraints)
+    if dry_run_task:
+        output = generate(prompt)
+        validation = validate_generated_output(output, bound_paths)
+        warnings = [] if validation.ok else [validation.reason]
+        return _task_result(target, prompt, output, applied=False, warnings=warnings)
+
     feedback = None
+    last_output = ""
+    last_validation = None
+    last_log = ""
     for t in range(1, MAX_ATTEMPTS + 1):
-        print(f"--- attempt {t} (provider={os.environ.get('SIMPLICIO_PROVIDER','claude')}) ---")
+        if not quiet:
+            print(f"--- attempt {t} (provider={os.environ.get('SIMPLICIO_PROVIDER','claude')}) ---")
         output = generate(prompt, feedback)
-        ok, log = _apply_and_test(output, root)
+        last_output = output or ""
+        last_validation = validate_generated_output(output, bound_paths)
+        ok, log = _apply_and_test(output, root, bound_paths)
+        last_log = log
         log_run(root, {
             "mode": "pipeline",
             "attempt": t,
@@ -92,9 +161,24 @@ def run(root, stack, goal, target, criteria, constraints):
             "stack": stack,
         })
         if ok:
-            print("PASSED the contract. DONE.")
-            return output
-        print("failed:", log[:300])
-        feedback = build_retry_feedback(t + 1, validate_generated_output(output), log)
-    print("attempts exhausted — manual review needed.")
+            if not quiet:
+                print("PASSED the contract. DONE.")
+            return _task_result(target, prompt, output, applied=True)
+        if not quiet:
+            print("failed:", log[:300])
+        feedback = build_retry_feedback(t + 1, last_validation, log)
+    if not quiet:
+        print("attempts exhausted — manual review needed.")
+    warnings = []
+    if last_validation and not last_validation.ok:
+        warnings.append(last_validation.reason)
+    elif last_log:
+        warnings.append(last_log[:500])
+    return _task_result(target, prompt, last_output, applied=False, warnings=warnings)
+
+def run(root, stack, goal, target, criteria, constraints, bound_paths=None):
+    result = run_task(root, stack, goal, target, criteria, constraints,
+                      bound_paths=bound_paths)
+    if result["applied"]:
+        return result
     return None

{simplicio_cli-0.4.1 → simplicio_cli-0.4.3/simplicio_cli.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: simplicio-cli
-Version: 0.4.1
+Version: 0.4.3
 Summary: Portable task-to-code pipeline that works with any LLM. Turn a one-line task into a verified code change — diff + test + verify loop. +55 pts on a 156-check benchmark, 21% faster, ~same tokens.
 Author-email: Wesley Simplicio <wesleybob4@gmail.com>
 License: MIT
@@ -31,8 +31,8 @@ Requires-Dist: sentence-transformers>=2.2
 Requires-Dist: numpy>=1.23
 Requires-Dist: anthropic>=0.30
 Requires-Dist: openai>=1.30
-Requires-Dist: simplicio-mapper>=0.6.0
-Requires-Dist: simplicio-prompt>=1.9.0
+Requires-Dist: simplicio-mapper>=0.6.1
+Requires-Dist: simplicio-prompt>=1.12.0
 Requires-Dist: httpx>=0.27
 Requires-Dist: orjson>=3.10
 Requires-Dist: diskcache>=5.6
@@ -134,12 +134,25 @@ M1 MacBook (8 GB), five sub-4B tiny models, six frontier 2026 models, and three
 mid-tier 7B–12B open models. Every one gained at least **+14 points** when
 wrapped in simplicio's 6-layer contract.
 
-#### Hugging Face — Qwen2.5-Coder, re-run on 2026-05-27 (latest mapper, 10 cases/side, 156 checks)
+#### Hugging Face — recommended Qwen3-Coder defaults (HF router)
 
-First batch of the smaller→larger re-benchmark against the latest
-`simplicio-mapper` artifacts. The 1.5B runs on CPU via `transformers`
-(Hugging Face Inference Providers does not serve it); the 3B and 7B run
-through the HF router (`https://router.huggingface.co/v1`).
+The served Qwen Coder recommendation now uses the Qwen3-Coder MoE family.
+`Qwen/Qwen2.5-Coder-3B-Instruct` and
+`Qwen/Qwen2.5-Coder-7B-Instruct` remain available as legacy fallback models for
+historical comparisons and hardware that cannot host the MoE successors.
+
+| Slot | Recommended model | Route | Notes |
+|---|---|---|---|
+| Efficient coder | `Qwen/Qwen3-Coder-30B-A3B-Instruct` | HF router | 30B total / ~3B active MoE successor to the 3B slot |
+| High-ceiling coder | `Qwen/Qwen3-Coder-Next` | HF router | 80B total / ~3B active MoE successor to the 7B slot |
+
+> Reproduce the new default set:
+> `BENCH_BASE_URL=https://router.huggingface.co/v1 BENCH_API_KEY=<hf-token>
+> BENCH_MODELS="Qwen/Qwen3-Coder-30B-A3B-Instruct,Qwen/Qwen3-Coder-Next"
+> python3 bench/run_offline.py`.
+
+Legacy Qwen2.5-Coder baseline, re-run on 2026-05-27 against the latest
+`simplicio-mapper` artifacts (10 cases/side, 156 checks):
 
 | Model | Without simplicio | With simplicio | Gain |
 |---|---|---|---|
@@ -148,10 +161,9 @@ through the HF router (`https://router.huggingface.co/v1`).
 | **Qwen 2.5 Coder 1.5B** (`Qwen/Qwen2.5-Coder-1.5B-Instruct`, local CPU) | 30% | **92%** | **+62 pts** |
 | **HF avg (3 models · 10 cases · 156 checks)** | **34%** | **94%** | **+60 pts (+172%)** |
 
-> Monotonic from smaller to larger: pass-rate with simplicio climbs **92% →
-> 94% → 96%** as the model grows, while the raw-prompt baseline stays at
-> **30–38%**. The 1.5B model gains the most (**+62 pts**) — the contract does
-> the heaviest lifting where the model is weakest. Reproduce:
+> Monotonic from smaller to larger in the legacy baseline: pass-rate with
+> simplicio climbs **92% → 94% → 96%** as the model grows, while the raw-prompt
+> baseline stays at **30–38%**. Reproduce the legacy set:
 > `BENCH_BASE_URL=https://router.huggingface.co/v1 BENCH_API_KEY=<hf-token>
 > BENCH_MODELS="local:Qwen/Qwen2.5-Coder-1.5B-Instruct,Qwen/Qwen2.5-Coder-3B-Instruct,Qwen/Qwen2.5-Coder-7B-Instruct"
 > python3 bench/run_offline.py`.
@@ -167,7 +179,18 @@ Pro) show `n/a` for the new column: their OpenRouter calls hit account-level
 HTTP 402 / provider failures on >50% of requests this round, so the sample is
 too small to publish; their old numbers still stand.
 
-#### Local offline — qwen2.5-coder on Ollama, M1 8 GB, run on 2026-05-27 (30 runs/side, 156 checks)
+#### Local offline — Qwen3-Coder GGUF recommendation, Qwen2.5 legacy baseline
+
+For local OpenAI-compatible servers, prefer the Qwen3-Coder GGUF builds when
+the machine can host MoE weights:
+
+| Slot | Recommended local weights | Notes |
+|---|---|---|
+| Efficient coder | `unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF` | Primary local successor for the 3B-active slot |
+| High-ceiling coder | `unsloth/Qwen3-Coder-Next-GGUF` | 24 GB GPU-class successor for long-context work |
+
+The last fully offline fallback baseline remains qwen2.5-coder on Ollama,
+M1 8 GB, run on 2026-05-27 (30 runs/side, 156 checks):
 
 | Model | Without simplicio | With simplicio | Gain |
 |---|---|---|---|
@@ -180,7 +203,7 @@ too small to publish; their old numbers still stand.
 > `http://localhost:11434/v1` (Ollama's OpenAI-compatible endpoint). A
 > 1.5B-param model running on a 4-year-old laptop reaches **88%** pass-rate
 > with simplicio's contract — same hardware, same model, raw prompt = 32%.
-> Reproduce: `BENCH_BASE_URL=http://localhost:11434/v1 BENCH_API_KEY=ollama
+> Reproduce the legacy fallback: `BENCH_BASE_URL=http://localhost:11434/v1 BENCH_API_KEY=ollama
 > BENCH_MODELS="qwen2.5-coder:7b" python3 bench/run_offline.py`.
 
 #### Tiny models — sub-4B, run on 2026-05-26 (50 runs/side, 260 checks)
@@ -382,6 +405,29 @@ simplicio task "..." --stack angular --target ...
 
 How it works: simplicio shells out to `claude -p "<prompt>"` (or `codex exec "<prompt>"`) as a subprocess, captures stdout, runs the test loop. The inner CLI authenticates via your existing OAuth session in `~/.claude/` or `~/.codex/`. simplicio sets `SIMPLICIO_HOOK_GUARD=1` in the subprocess env so the inner Claude Code session does **not** re-fire simplicio's own UserPromptSubmit hook (no infinite recursion).
 
+For orchestrators such as SendSprint, `simplicio task` also has a structured
+contract:
+
+```bash
+simplicio task "hide Delete button for non-admins" \
+  --stack angular \
+  --target src/app/screen/screen.component.html \
+  --dry-run-task \
+  --json
+
+simplicio task "front-only task" \
+  --stack angular \
+  --target src/app/screen/screen.component.html \
+  --bound-paths "src/app/**" \
+  --json
+```
+
+`--dry-run-task` generates the would-be diff/test output without applying or
+testing it. `--json` returns `{task_id, applied, files_changed, tokens_used,
+cost_usd, diff_summary, warnings}`. Repeat `--bound-paths <glob>` to reject
+diffs outside the allowed edit surface; violations are reported in `warnings`
+and the command exits non-zero.
+
 ### Path 3 example — standalone with API key
 
 ```bash

{simplicio_cli-0.4.1 → simplicio_cli-0.4.3}/simplicio_cli.egg-info/requires.txt

@@ -2,8 +2,8 @@ sentence-transformers>=2.2
 numpy>=1.23
 anthropic>=0.30
 openai>=1.30
-simplicio-mapper>=0.6.0
-simplicio-prompt>=1.9.0
+simplicio-mapper>=0.6.1
+simplicio-prompt>=1.12.0
 httpx>=0.27
 orjson>=3.10
 diskcache>=5.6

simplicio_cli-0.4.1/simplicio/__init__.py

@@ -1 +0,0 @@
-__version__ = "0.1.0"