chad-code 1.0.2__tar.gz → 1.0.4__tar.gz

{chad_code-1.0.2/src/chad_code.egg-info → chad_code-1.0.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: chad-code
-Version: 1.0.2
+Version: 1.0.4
 Summary: Local MLX-backed, Claude-Code-style coding agent (Apple Silicon, Ornith 35B/9B)
 License-Expression: MIT
 Project-URL: Repository, https://github.com/nathansutton/chad
@@ -39,7 +39,7 @@ Dynamic: license-file
 
 [![tests](https://github.com/nathansutton/chad/actions/workflows/tests.yml/badge.svg)](https://github.com/nathansutton/chad/actions/workflows/tests.yml)
 
-<img src="docs/tbench-size-vs-score.png" width="840" alt="Terminal-Bench 2.0: accuracy vs. model size. chad + Ornith (a 35B MoE) lands on the Claude Sonnet 4.5 line, matching open models many times its size and standing alone in the laptop-class ≤40B band.">
+<img src="docs/tbench-size-vs-score.png" width="840" alt="Terminal-Bench 2.1: accuracy vs. cost per run. Every verified entry is a proprietary frontier model in a datacenter, costing $130–$2,000 per run. chad + Ornith (a 35B MoE) clears 57% on an Apple Silicon laptop for the electricity — the only no-API-cost point on the board.">
 
 > Claude can do anything, for anyone, anywhere. chad does one thing. 🗿
 > *Coding under supervision.*
@@ -86,20 +86,20 @@ instrument:
 > Real session, unedited (cold model load cut): a local 35B reasons through the failure,
 > edits the file, reruns the tests, confirms green.
 
-## Sonnet on your laptop
+## Frontier scores, laptop cost
 
 The exam is [Terminal-Bench](https://www.tbench.ai/leaderboard), the standard benchmark for
-CLI coding agents. chad won't top it — the A students are frontier models in datacenters.
-The number worth looking at is **how much capability chad wrings out of 35B parameters**:
-on Terminal-Bench 2.1, chad + Ornith lands around the **Claude Sonnet 4.5** line, matching
-open models many times its size and beating every open model in its own weight class by a
-wide margin. On a laptop, *capability per parameter* is the axis you actually compete on —
-and that's where chad + Ornith is out on the frontier alone.
-
-> **Placeholder number** — the full verified run is still in flight (`≈50%`, k=1). The whole
-> benchmark is **publicly reproducible from a Mac**: the exact Harbor adapter, the runner,
-> and the recipe live in [`benchmarks/tb2/`](benchmarks/tb2/README.md). Check it, don't
-> trust it.
+CLI coding agents. chad won't top it — every verified entry on Terminal-Bench 2.1 is a
+frontier model in a datacenter, scoring 59–84%. The number worth looking at is what that
+capability **costs**: the paid field spends **$130–$2,000 in API fees per run**. chad +
+Ornith clears **57%** on an Apple Silicon laptop, for the electricity — no API, no tokens,
+no datacenter. On a laptop, *capability per dollar* is the axis you actually compete on, and
+chad is the only point on the board that runs there.
+
+> **Provisional number** — `57%` (51/89, k=1, self-run, not yet leaderboard-verified). The
+> whole benchmark is **publicly reproducible from a Mac**: the exact Harbor adapter, the
+> runner, and the recipe live in [`benchmarks/tb2/`](benchmarks/tb2/README.md). Check it,
+> don't trust it.
 
 ## The bet: at this end of the report card, the harness beats the model
 
@@ -214,7 +214,7 @@ Both are covered in full in the [Configuration reference](docs/configuration.md)
   cache, the trimmable/append-only trade, and the ideas borrowed from other agents.
 - **[Throughput & performance](docs/benchmarks.md)** — prefill / decode / warm-step numbers
   you can reproduce with `chad-bench`.
-- **[Terminal-Bench 2.0 reproduction](benchmarks/tb2/README.md)** — the exact Harbor adapter
+- **[Terminal-Bench 2.1 reproduction](benchmarks/tb2/README.md)** — the exact Harbor adapter
   and runner behind the chart; serve Ornith yourself and check the number.
 - **[Configuration reference](docs/configuration.md)** — Agent Skills, MCP servers, the
   context window, every environment variable, and the safety opt-outs.

{chad_code-1.0.2 → chad_code-1.0.4}/README.md

@@ -2,7 +2,7 @@
 
 [![tests](https://github.com/nathansutton/chad/actions/workflows/tests.yml/badge.svg)](https://github.com/nathansutton/chad/actions/workflows/tests.yml)
 
-<img src="docs/tbench-size-vs-score.png" width="840" alt="Terminal-Bench 2.0: accuracy vs. model size. chad + Ornith (a 35B MoE) lands on the Claude Sonnet 4.5 line, matching open models many times its size and standing alone in the laptop-class ≤40B band.">
+<img src="docs/tbench-size-vs-score.png" width="840" alt="Terminal-Bench 2.1: accuracy vs. cost per run. Every verified entry is a proprietary frontier model in a datacenter, costing $130–$2,000 per run. chad + Ornith (a 35B MoE) clears 57% on an Apple Silicon laptop for the electricity — the only no-API-cost point on the board.">
 
 > Claude can do anything, for anyone, anywhere. chad does one thing. 🗿
 > *Coding under supervision.*
@@ -49,20 +49,20 @@ instrument:
 > Real session, unedited (cold model load cut): a local 35B reasons through the failure,
 > edits the file, reruns the tests, confirms green.
 
-## Sonnet on your laptop
+## Frontier scores, laptop cost
 
 The exam is [Terminal-Bench](https://www.tbench.ai/leaderboard), the standard benchmark for
-CLI coding agents. chad won't top it — the A students are frontier models in datacenters.
-The number worth looking at is **how much capability chad wrings out of 35B parameters**:
-on Terminal-Bench 2.1, chad + Ornith lands around the **Claude Sonnet 4.5** line, matching
-open models many times its size and beating every open model in its own weight class by a
-wide margin. On a laptop, *capability per parameter* is the axis you actually compete on —
-and that's where chad + Ornith is out on the frontier alone.
-
-> **Placeholder number** — the full verified run is still in flight (`≈50%`, k=1). The whole
-> benchmark is **publicly reproducible from a Mac**: the exact Harbor adapter, the runner,
-> and the recipe live in [`benchmarks/tb2/`](benchmarks/tb2/README.md). Check it, don't
-> trust it.
+CLI coding agents. chad won't top it — every verified entry on Terminal-Bench 2.1 is a
+frontier model in a datacenter, scoring 59–84%. The number worth looking at is what that
+capability **costs**: the paid field spends **$130–$2,000 in API fees per run**. chad +
+Ornith clears **57%** on an Apple Silicon laptop, for the electricity — no API, no tokens,
+no datacenter. On a laptop, *capability per dollar* is the axis you actually compete on, and
+chad is the only point on the board that runs there.
+
+> **Provisional number** — `57%` (51/89, k=1, self-run, not yet leaderboard-verified). The
+> whole benchmark is **publicly reproducible from a Mac**: the exact Harbor adapter, the
+> runner, and the recipe live in [`benchmarks/tb2/`](benchmarks/tb2/README.md). Check it,
+> don't trust it.
 
 ## The bet: at this end of the report card, the harness beats the model
 
@@ -177,7 +177,7 @@ Both are covered in full in the [Configuration reference](docs/configuration.md)
   cache, the trimmable/append-only trade, and the ideas borrowed from other agents.
 - **[Throughput & performance](docs/benchmarks.md)** — prefill / decode / warm-step numbers
   you can reproduce with `chad-bench`.
-- **[Terminal-Bench 2.0 reproduction](benchmarks/tb2/README.md)** — the exact Harbor adapter
+- **[Terminal-Bench 2.1 reproduction](benchmarks/tb2/README.md)** — the exact Harbor adapter
   and runner behind the chart; serve Ornith yourself and check the number.
 - **[Configuration reference](docs/configuration.md)** — Agent Skills, MCP servers, the
   context window, every environment variable, and the safety opt-outs.

{chad_code-1.0.2 → chad_code-1.0.4}/pyproject.toml

@@ -4,7 +4,7 @@
 # import name, and command name are independent. `uvx chad-code` runs the alias
 # script added under [project.scripts].
 name = "chad-code"
-version = "1.0.2"
+version = "1.0.4"
 description = "Local MLX-backed, Claude-Code-style coding agent (Apple Silicon, Ornith 35B/9B)"
 readme = "README.md"
 license = "MIT"

{chad_code-1.0.2 → chad_code-1.0.4}/src/chad/agent.py

@@ -435,6 +435,11 @@ class Agent:
         self.draft_accepted = 0
         self.think_tokens = 0   # tokens spent inside <think> blocks (reasoning overhead)
         self.think_capped = 0   # times the soft think-cap force-closed a step
+        # Real tool dispatches this agent has executed (fn actually ran — excludes
+        # terminal `done`, validation rejects, and harness-injected nudge messages,
+        # which reuse real tool names in the transcript). The evidence signal behind
+        # the sub-agent zero-evidence warning in _run_subagent.
+        self.tool_dispatches = 0
         # prefill accounting: the master cost for a local model is how many *new*
         # tokens it has to prefill across a turn (context bloat -> big prefills).
         # This is the metric symbolic/repo-map retrieval is meant to shrink.
@@ -689,12 +694,22 @@ class Agent:
             elif (not result or result.startswith("[stopped:")
                     or result.startswith("[task failed:")):
                 result = _salvage(result or "[task returned nothing]")
-        # tool_calls (not `forwards`, a speculative-decoding counter that is 0 on the
-        # normal path) is the number that diagnoses a sub-agent returning nothing: it
-        # separates "never got to search" from "searched and lost its findings".
-        log.info("TASK end | desc=%r | %.1fs | tool_calls=%d gen=%d prefill=%d | -> %s",
+            else:
+                # A confident, non-empty report produced with ZERO tool dispatches came
+                # from model memory, not this repo — the one sub-agent failure the empty/
+                # crashed salvage above cannot see, and the most dangerous fold-back: it
+                # reads as evidence (guardrails.subagent_evidence_warning).
+                warned = guardrails.subagent_evidence_warning(result, sub.tool_dispatches)
+                if warned is not None:
+                    log.info("TASK zero-evidence warning appended | desc=%r", description)
+                    result = warned
+        # tool_dispatches (not a transcript count — harness nudges reuse real tool names
+        # in tool-role messages) is the number that diagnoses a sub-agent returning
+        # nothing: it separates "never got to search" from "searched and lost its
+        # findings", and 0 with a confident report is the answered-from-memory tell.
+        log.info("TASK end | desc=%r | %.1fs | tool_dispatches=%d gen=%d prefill=%d | -> %s",
                  description, time.perf_counter() - _t0,
-                 sum(1 for m in sub.messages if m.get("role") == "tool") if sub else 0,
+                 sub.tool_dispatches if sub else 0,
                  sub.gen_tokens if sub else 0,
                  sub.prefill_tokens if sub else 0, result_preview(result or ""))
         return result or "[task returned nothing]"
@@ -786,6 +801,13 @@ class Agent:
         gate_nudges = 0       # times the investigation->edit gate fired this turn
         subagent_sigs = set() # (description, prompt) of sub-agents already spawned this turn
         truncation_nudges = 0  # times we pushed past a token-cap truncation this turn
+        garble_nudges = 0  # malformed-tool-call re-nudges (own counter, so a step-0
+                           # cap-hit can't spend the garble recovery budget)
+        consecutive_garbles = 0  # back-to-back garbled steps (drives exemplar + scrub)
+        last_garble_idx = None   # self.messages index of the previous garbled assistant
+                                 # message (scrubbed if the next step garbles too)
+        done_audit_bounces = 0   # total audit bounces this turn (hard cap 2)
+        audit_absent_list = []   # task-named paths the FIRST audit stat'ed as absent
         landing_nudges = 0  # one-shot "you're out of steps, land the edit" near the cap
         consecutive_failed_bash = 0  # back-to-back errored bash with no edit (thrash)
         thrash_nudges = 0
@@ -1400,12 +1422,33 @@ class Agent:
                 _stripped_for_markers = strip_think(text)
                 garbled = ("<tool_call>" in _stripped_for_markers
                            or "</function>" in _stripped_for_markers
-                           or "<function=" in _stripped_for_markers)
+                           or "<function=" in _stripped_for_markers
+                           or _has_open_tool_call(text))
+                if garbled:
+                    consecutive_garbles += 1
+                    # From the 2nd consecutive garble, scrub the PREVIOUS garbled
+                    # assistant body: the wrong dialect must not stay in context as a
+                    # few-shot example (repeated garbles in a row each condition on the
+                    # last). Costs one prefix-cache invalidation on this rare path.
+                    if (levers.enabled("garble_never_final")
+                            and consecutive_garbles >= 2 and last_garble_idx is not None):
+                        self.messages[last_garble_idx]["content"] = \
+                            guardrails.GARBLE_SCRUBBED
+                        log.info("GARBLE scrub at step %d: previous garbled message "
+                                 "body removed from context (consecutive=%d)",
+                                 step, consecutive_garbles)
+                    last_garble_idx = len(self.messages) - 1
+                else:
+                    consecutive_garbles = 0
                 kind, nudge = guardrails.nudge_for_no_calls(
                     text, hit_cap, made_edit, unverified_edit, read_only_intent,
                     action_task or run_task, truncation_nudges, answer_nudges,
-                    verify_nudges, _has_open_tool_call(text), garbled_call=garbled)
-                if kind == "truncated":
+                    verify_nudges, _has_open_tool_call(text), garbled_call=garbled,
+                    garble_nudges=garble_nudges,
+                    consecutive_garbles=consecutive_garbles)
+                if kind == "garble":
+                    garble_nudges += 1
+                elif kind == "truncated":
                     truncation_nudges += 1
                 elif kind == "no-edit":
                     answer_nudges += 1
@@ -1417,6 +1460,21 @@ class Agent:
                              has_code, action_task, run_task)
                     self.messages.append({"role": "tool", "name": "edit", "content": nudge})
                     continue
+                # Invariant: a step whose text contains tool-call markers is NEVER a
+                # final answer — the model was trying to act, not to finish. With the
+                # nudge budget spent, hard-stop with a banked note instead of accepting
+                # the garble (otherwise a run of garbles can burn the shared counter and
+                # the audit latch, and the last one ships as the answer with most of the
+                # wall budget still unspent).
+                if garbled and levers.enabled("garble_never_final"):
+                    self.budget_note = guardrails.progress_note(self.messages)
+                    log.info("END step %d: GARBLE hard-stop (%d garble nudges spent) — "
+                             "a garbled tool call is never a final answer",
+                             step, garble_nudges)
+                    self._emit("info", "  [turn stopped: repeated malformed tool calls "
+                                       "— progress note banked; say 'continue' to retry]")
+                    return ("[stopped: the model kept emitting malformed tool calls "
+                            "— say 'continue' to resume]")
                 # Iter-3 did-nothing gate: in auto/headless mode (every benchmark run), a
                 # turn that ends having executed ZERO real tools is never a legitimate
                 # completion — the keyword intent classifier misses tasks like "extract the
@@ -1465,6 +1523,8 @@ class Agent:
                     })
                     if audit:
                         done_audit_fired = True
+                        done_audit_bounces += 1
+                        audit_absent_list = guardrails.audit_absent_paths(audit_task)
                         _runway = ((self._turn_budget_s
                                     - (time.monotonic() - turn_start))
                                    if self._turn_budget_s else float("inf"))
@@ -1474,6 +1534,26 @@ class Agent:
                         self.messages.append({"role": "tool", "name": "edit",
                                               "content": audit})
                         continue
+                # Absent-path re-bounce: the first audit stat'ed task-named paths as
+                # ABSENT and promised acceptance anyway; if they are STILL absent with
+                # runway to spare, bounce one final time (cap 2/turn) — otherwise a
+                # task-named deliverable the audit already flagged absent can go
+                # unwritten and the accept still goes through.
+                if self.mode != "plan" and not self._subagent \
+                        and done_audit_fired and done_audit_bounces < 2 \
+                        and audit_absent_list:
+                    _runway = ((self._turn_budget_s
+                                - (time.monotonic() - turn_start))
+                               if self._turn_budget_s else None)
+                    rebounce = guardrails.audit_rebounce(audit_absent_list, _runway)
+                    if rebounce:
+                        done_audit_bounces += 1
+                        audit_absent_list = []
+                        log.info("DONE-AUDIT rebounce (final-answer): still-absent "
+                                 "task paths, runway=%s", _runway)
+                        self.messages.append({"role": "tool", "name": "edit",
+                                              "content": rebounce})
+                        continue
                 log.info("END step %d: model produced a FINAL ANSWER, no tool calls "
                          "(did_work=%s, made_edit=%s, unverified_edit=%s)",
                          step, did_work, made_edit, unverified_edit)
@@ -1483,6 +1563,8 @@ class Agent:
             # The step acted — the spiral broke on its own, so clear the no-think escalation
             # streak: escalation is only for consecutive capped/salvaged stalls.
             capped_stall_streak = 0
+            consecutive_garbles = 0  # a parsed call ends the garble streak
+            last_garble_idx = None
 
             # Terminal tool -> end the turn cleanly, but enforce verify-before-done
             # (forge's prerequisite idea): if files were changed and nothing has been
@@ -1574,6 +1656,8 @@ class Agent:
                         })
                         if audit:
                             done_audit_fired = True
+                            done_audit_bounces += 1
+                            audit_absent_list = guardrails.audit_absent_paths(audit_task)
                             _runway = ((self._turn_budget_s
                                         - (time.monotonic() - turn_start))
                                        if self._turn_budget_s else float("inf"))
@@ -1582,6 +1666,22 @@ class Agent:
                             self.messages.append({"role": "tool", "name": "done",
                                                   "content": audit})
                             continue
+                    # Absent-path re-bounce (see the final-answer twin above): a
+                    # still-absent task-named path at accept time gets ONE more bounce,
+                    # then the next done is accepted unconditionally.
+                    if done_audit_fired and done_audit_bounces < 2 and audit_absent_list:
+                        _runway = ((self._turn_budget_s
+                                    - (time.monotonic() - turn_start))
+                                   if self._turn_budget_s else None)
+                        rebounce = guardrails.audit_rebounce(audit_absent_list, _runway)
+                        if rebounce:
+                            done_audit_bounces += 1
+                            audit_absent_list = []
+                            log.info("DONE-AUDIT rebounce: still-absent task paths, "
+                                     "runway=%s", _runway)
+                            self.messages.append({"role": "tool", "name": "done",
+                                                  "content": rebounce})
+                            continue
                 # Iter-3 deliverable recheck (levers.done_spec_recheck): one last self-check
                 # that the required outputs actually exist at the right path/format before
                 # we accept done — the hidden container-end-state verifier gives no second
@@ -1725,6 +1825,7 @@ class Agent:
                     result = "[denied by user]"
                 else:
                     _t0 = time.perf_counter()
+                    self.tool_dispatches += 1
                     try:
                         result = fn(args, self._should_stop)
                         if plan_write and result.startswith("[wrote"):
@@ -1736,6 +1837,15 @@ class Agent:
                     # whole — several calls in one step stack into one prefill, so later
                     # results only get what's left of the step budget (floor-protected).
                     result = _clip_tool_result(result, cap=_step_tool_cap(step_tool_chars))
+                    # Duplicate read-only output: if this (clipped) result is
+                    # byte-identical to a tool message still in the transcript, append a
+                    # short pointer instead of paying the body's prefill again — see
+                    # guardrails.elide_duplicate_result for the safety argument.
+                    _elided = guardrails.elide_duplicate_result(name, result, self.messages)
+                    if _elided is not None:
+                        log.info("TOOL %s duplicate result elided (%d chars)",
+                                 name, len(result))
+                        result = _elided
                     step_tool_chars += len(result)
                     if _PREFILL_TRACE:
                         self._trace_tools_pending.append([name, round(_tool_s, 4)])
@@ -1854,7 +1964,18 @@ class Agent:
             if made_edit and not _gov_prev_made:
                 readonly_streak = 0
             elif did_work and not made_edit:
-                readonly_streak += 1
+                # A bash step that isn't provably read-only is ACTION, not
+                # investigation — `git merge`, `apt-get install`, redirects are ops
+                # progress. Otherwise the gate can count an entire git/ops workflow as
+                # "investigation" and demand an edit at a decision point where there is
+                # nothing to edit yet.
+                if levers.enabled("gate_ops_exempt") and any(
+                        n == "bash"
+                        and not guardrails.is_readonly_bash(str(a.get("command", "")))
+                        for n, a in calls):
+                    readonly_streak = 0
+                else:
+                    readonly_streak += 1
             if not read_only_intent:
                 gate = guardrails.investigation_gate(readonly_streak, made_edit, gate_nudges)
                 if gate:

{chad_code-1.0.2 → chad_code-1.0.4}/src/chad/config.py

@@ -31,6 +31,21 @@ def eq(name: str, expected: str) -> bool:
     return os.environ.get(name) == expected
 
 
+def traces_enabled() -> bool:
+    """Whether chad may write local diagnostic traces under ~/.chad — the readable
+    `session.log` (user query, tool-call args, bash/write/edit previews, result previews)
+    and the persistent input history at ~/.chad/history.
+
+    Privacy-first default: **OFF**. chad is a local, single-user agent, so these traces
+    never leave the machine, but they still record command/file previews in plaintext
+    outside the repo — so they are opt-in. Set **`CHAD_SESSION_LOG`** (any truthy value)
+    to turn them on. `CHAD_NO_SESSION_LOG`, if set, forces them OFF and wins over the
+    opt-in (a hard kill switch, kept for compatibility with the pre-opt-in default)."""
+    if flag("CHAD_NO_SESSION_LOG"):
+        return False
+    return flag("CHAD_SESSION_LOG")
+
+
 def env_str(name, default=None):
     """The env value if set to a non-empty string, else `default`. Mirrors the common
     `os.environ.get(name) or <fallback>` idiom (empty string collapses to the default)."""

{chad_code-1.0.2 → chad_code-1.0.4}/src/chad/diag.py

@@ -9,8 +9,10 @@ declare done, which cache served the turn. This module is that trace.
 
 The log is bounded (5 MB x3 rotation) and previews pass through a best-effort secret
 redactor, but it still records command/file previews in plaintext outside the repo, so
-treat it as sensitive. None of this touches the model-facing transcript or the tool
-results the model sees — it is the diagnostic trace only.
+treat it as sensitive. For that reason it is **opt-in**: privacy-first, chad writes no
+trace under ~/.chad unless CHAD_SESSION_LOG is set (see config.traces_enabled). None of
+this touches the model-facing transcript or the tool results the model sees — it is the
+diagnostic trace only.
 """
 import json
 import logging
@@ -22,11 +24,12 @@ from . import config
 
 _LOG_DIR = os.path.expanduser("~/.chad")
 log = logging.getLogger("chad")
-# Local privacy opt-out: set CHAD_NO_SESSION_LOG (any truthy value) to disable the
-# diagnostic session log entirely, matching the CHAD_NO_VALIDATE convention. When opted
-# out we install a NullHandler (so the many log.info calls stay cheap no-ops and Python
-# never warns about missing handlers) and never create ~/.chad for the log's sake.
-_DISABLED = config.flag("CHAD_NO_SESSION_LOG")
+# Privacy-first default: the diagnostic session log is OFF unless opted in. Set
+# CHAD_SESSION_LOG (any truthy value) to enable it; CHAD_NO_SESSION_LOG still forces it
+# off (see config.traces_enabled). When disabled we install a NullHandler (so the many
+# log.info calls stay cheap no-ops and Python never warns about missing handlers) and
+# never create ~/.chad for the log's sake.
+_DISABLED = not config.traces_enabled()
 if _DISABLED:
     log.addHandler(logging.NullHandler())
     log.propagate = False