unwedge 0.1.0__tar.gz

unwedge-0.1.0/.agents/plugins/marketplace.json

@@ -0,0 +1,20 @@
+{
+  "name": "unwedge",
+  "plugins": [
+    {
+      "name": "unwedge",
+      "source": {
+        "source": "local",
+        "path": "./plugins/codex"
+      },
+      "policy": {
+        "installation": "AVAILABLE",
+        "authentication": "ON_INSTALL"
+      },
+      "category": "Productivity",
+      "interface": {
+        "displayName": "unwedge circuit breaker"
+      }
+    }
+  ]
+}

unwedge-0.1.0/.claude-plugin/marketplace.json

@@ -0,0 +1,21 @@
+{
+  "name": "unwedge",
+  "owner": {
+    "name": "umithavare",
+    "url": "https://github.com/umithavare"
+  },
+  "metadata": {
+    "description": "unwedge: a circuit breaker for AI agent doom loops.",
+    "version": "0.1.0"
+  },
+  "plugins": [
+    {
+      "name": "unwedge",
+      "source": "./plugins/claude-code",
+      "description": "Spots doom loops (repeated failing commands, rereading, stalled progress) and logs them or hints Claude. Needs the unwedge CLI: uv tool install unwedge.",
+      "version": "0.1.0",
+      "license": "Apache-2.0",
+      "keywords": ["doom-loop", "circuit-breaker", "cost", "hooks"]
+    }
+  ]
+}

unwedge-0.1.0/.gitignore

@@ -0,0 +1,27 @@
+# Python
+__pycache__/
+*.py[cod]
+*.egg-info/
+build/
+dist/
+.venv/
+.pytest_cache/
+.ruff_cache/
+.mypy_cache/
+htmlcov/
+.coverage
+
+# benchmark data (re-fetch with benchmarks/fetch_trajectories.py) and raw per-turn outputs
+benchmarks/data/
+benchmarks/results/*.jsonl
+benchmarks/results/*.log
+
+# local model weights and servers
+models/
+.laya/
+
+# editors / OS
+.idea/
+.vscode/
+.DS_Store
+Thumbs.db

unwedge-0.1.0/CHANGELOG.md

@@ -0,0 +1,20 @@
+# Changelog
+
+## 0.1.0 — 2026-09-23
+
+First public version.
+
+- Guard with a code tier (repeats, result fingerprints, error signatures, streaks) and an optional
+  typed-decision provider; code-first policy with hysteresis, cooldowns and hint-before-stop.
+  Two hints, then an escalation that repeats at most every eight turns; the provider only picks
+  which hint, never how strong the intervention is.
+- Providers: TypeSafe jev (hosted), Laya via `laya-serve` or `unwedge serve` (local, open weights,
+  laya-mlx on Apple silicon), in-process Laya for the Python API, or none. Zero-dependency HTTP
+  client; every provider failure fails open.
+- Claude Code plugin and marketplace (background hooks; hint and stop modes).
+- Codex CLI support through `hooks.json`; experimental Codex plugin marketplace.
+- `unwedge hook | scan | replay | report | serve | doctor`.
+- A tool result that talks to the reader (a "gate" firing) vetoes the provider's judgments for
+  that window and is logged; code-tier alarms still act, and nothing is sent to the agent.
+- Benchmark on 218 public SWE-agent trajectories (code only vs jev, plus paired Laya
+  comparisons), with reproducible scripts.

unwedge-0.1.0/CONTRIBUTING.md

@@ -0,0 +1,36 @@
+# Contributing
+
+Thanks for helping. unwedge is small on purpose: a zero-dependency core, one hook command, and a
+benchmark that keeps us honest.
+
+## Setup
+
+```bash
+git clone https://github.com/umithavare/unwedge && cd unwedge
+python -m venv .venv && . .venv/bin/activate      # Windows: .venv\Scripts\activate
+pip install -e ".[dev,bench]"
+ruff check src tests benchmarks && pytest -q
+```
+
+## Ground rules
+
+- **The core stays dependency-free.** It runs in a hook process on every tool call.
+- **The guard never breaks the agent.** Provider or hook failures fail open; add a test when you
+  touch that path.
+- **Code counts, models judge.** Anything code can compute exactly (repeats, counts, time,
+  budgets) stays in code. Providers only answer questions code cannot.
+- **Nothing from tool output reaches the agent.** Hints are human-written; providers only select one.
+- **Numbers come from the benchmark.** If you change the policy, questions or digest, rerun
+  `benchmarks/analyze.py` and update the README tables with what it prints, good or bad.
+
+## Adding a harness
+
+Write an adapter that turns the harness's tool-call payload into a `Turn`
+(`src/unwedge/adapters/`), teach `hooks.detect_harness` to recognize it, and add tests with
+real payload samples.
+
+## Adding a provider
+
+Implement the `Provider` protocol (`src/unwedge/providers/base.py`) or expose a
+`POST /v1/systemone` endpoint and use the `laya` provider with `LAYA_URL`. Add a `Profile` that
+matches the model's context size (full or compact battery).

unwedge-0.1.0/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

unwedge-0.1.0/NOTICE

@@ -0,0 +1,12 @@
+unwedge (unwedge)
+Copyright 2026 umithavare
+
+This product is licensed under the Apache License, Version 2.0 (see LICENSE).
+
+The benchmark samples trajectories from the public Hugging Face dataset
+nebius/SWE-agent-trajectories; the data is downloaded by benchmarks/fetch_trajectories.py
+and is not redistributed in this repository.
+
+TypeSafe and jev are products of TypeSafe AI. Laya is by Convai Innovations; laya-mlx is an
+independent MLX port. This project is independent and not affiliated with or endorsed by any
+of them.

unwedge-0.1.0/PKG-INFO

@@ -0,0 +1,212 @@
+Metadata-Version: 2.5
+Name: unwedge
+Version: 0.1.0
+Summary: A circuit breaker for AI coding agents: catches doom loops in Claude Code, Codex CLI and custom agent loops, with optional judgments from TypeSafe jev or a local Laya model.
+Project-URL: Homepage, https://github.com/umithavare/unwedge
+Project-URL: Documentation, https://github.com/umithavare/unwedge/tree/main/docs
+Project-URL: Issues, https://github.com/umithavare/unwedge/issues
+Author: umithavare
+License-Expression: Apache-2.0
+License-File: LICENSE
+License-File: NOTICE
+Keywords: ai-agents,circuit-breaker,claude-code,codex,doom-loop,hooks,jev,laya,llm-cost,typed-decisions
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Developers
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3 :: Only
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Software Development :: Quality Assurance
+Requires-Python: >=3.10
+Provides-Extra: bench
+Requires-Dist: httpx>=0.27; extra == 'bench'
+Requires-Dist: matplotlib>=3.7; extra == 'bench'
+Requires-Dist: numpy>=1.24; extra == 'bench'
+Requires-Dist: scikit-learn>=1.3; extra == 'bench'
+Provides-Extra: dev
+Requires-Dist: numpy>=1.24; extra == 'dev'
+Requires-Dist: pytest>=8; extra == 'dev'
+Requires-Dist: ruff>=0.6; extra == 'dev'
+Provides-Extra: laya
+Requires-Dist: laya>=0.3.11; extra == 'laya'
+Provides-Extra: laya-mlx
+Requires-Dist: laya-mlx>=0.2; (sys_platform == 'darwin' and platform_machine == 'arm64') and extra == 'laya-mlx'
+Description-Content-Type: text/markdown
+
+# unwedge
+
+**A circuit breaker for AI coding agents.** unwedge notices when Claude Code, Codex CLI or your
+own agent loop is stuck (the same failing command again, the same error ignored, no progress
+toward the task) and says so before the loop burns your budget.
+
+*Wedged* is old developer slang for a process that is stuck and cannot go on without help;
+unwedge spots a wedged agent, nudges it, and tells you when a nudge is not enough.
+
+[![ci](https://github.com/umithavare/unwedge/actions/workflows/ci.yml/badge.svg)](https://github.com/umithavare/unwedge/actions/workflows/ci.yml)
+[![license](https://img.shields.io/badge/license-Apache--2.0-blue.svg)](LICENSE)
+![python](https://img.shields.io/badge/python-3.10%2B-blue.svg)
+
+[Türkçe](README.tr.md) · [Claude Code](docs/claude-code.md) · [Codex](docs/codex.md) ·
+[Providers](docs/providers.md) · [How it works](docs/how-it-works.md) · [Benchmark](docs/benchmark.md)
+
+---
+
+Agents fail in a characteristic way: stuck at turn 9, still retrying at turn 60. `max_turns`,
+token caps and timeouts protect the budget, not the behaviour, so they fire after the money is
+spent. unwedge watches the behaviour:
+
+```text
+T6  edit 199:205 (+19 lines) -> edit REJECTED, not applied: F821 undefined name   S=0.97 P0=0.13
+T7  edit 199:205 (+19 lines) -> edit REJECTED, not applied: F821 undefined name   S=0.97 P0=0.18
+T8  edit 199:205 (+19 lines) -> edit REJECTED, not applied: F821 undefined name   S=0.97 P0=0.19  <<< hint
+    [UNWEDGE] You have run this command 4 times with the same result. Change something before
+    running it again, or step back and re-read the error.
+T11 ...                                                                                            <<< hint
+T14 ...                                                                                            <<< escalate
+```
+
+*A real SWE-agent session from the benchmark, replayed with live jev judgments (S = stall,
+P0 = probability of no progress). The agent's edit was rejected 31 times in a row, and 94% of
+the session's spend came after unwedge's first hint.
+[Full replay](benchmarks/results/replay_larpix.txt).*
+
+## What it does
+
+- **Records every tool call** through a hook and computes loop signals in code: repeated actions
+  with nothing changed in between, repeated results, repeated error signatures, streaks. Free and local.
+- **Optionally asks a typed-decision model** the questions code cannot answer, relative to *this*
+  session's task: is the agent still getting closer, is the work already done, is it drifting.
+  Providers: [TypeSafe jev](https://docs.typesafe.ai) (hosted) or
+  [Laya](https://github.com/NandhaKishorM/laya) (open weights, runs locally;
+  [laya-mlx](https://github.com/mizorewww/laya-mlx) on Apple silicon). Models only return
+  probabilities; they never write text to your agent.
+- **Acts according to its mode:** `shadow` only records; `hint` adds a short, human-written hint to
+  the agent's context; `stop` (opt-in) can end a session that keeps looping after two hints.
+- **Never breaks the agent:** every provider or hook problem fails open.
+
+## Quick start
+
+```bash
+uv tool install git+https://github.com/umithavare/unwedge   # PyPI (soon): uv tool install unwedge
+unwedge scan        # replay your recent Claude Code and Codex sessions and flag loops; free, local
+unwedge doctor      # check settings and the provider
+```
+
+**Claude Code**
+
+```text
+/plugin marketplace add umithavare/unwedge
+/plugin install unwedge@unwedge
+```
+
+The plugin runs in the background (no added latency) and asks for its mode and provider when you
+enable it. Details, manual `settings.json` setup and stop mode: [docs/claude-code.md](docs/claude-code.md).
+
+**Codex CLI** (0.124+): add to `~/.codex/hooks.json`, then approve it in `/hooks`:
+
+```json
+{ "hooks": { "PostToolUse": [{ "hooks": [{ "type": "command", "command": "unwedge hook", "timeout": 15 }] }] } }
+```
+
+More, including the plugin marketplace: [docs/codex.md](docs/codex.md).
+
+**Your own agent loop**
+
+```python
+from unwedge import Guard, GuardConfig, make_provider
+from unwedge.adapters.generic import turn_from_command
+
+provider = make_provider("none")                      # or "jev", "laya"
+guard = Guard(goal=task, provider=provider, config=GuardConfig.for_provider(provider, shadow=False))
+outcome = guard.on_turn(turn_from_command(index, command, output))
+if outcome.hint_text:
+    next_message += outcome.hint_text
+```
+
+See [docs/python-api.md](docs/python-api.md).
+
+## Providers
+
+| | `none` (default) | `jev` | `laya` |
+|---|---|---|---|
+| runs | locally, code only | TypeSafe API (US) | your machine (`laya-serve` or `unwedge serve`) |
+| cost per judged turn | free | ~$0.0001 | free |
+| latency per judged turn | none | p50 0.32 s, p99 0.64 s | p50 2.4 s (`multilingual`) to 4.1 s (`english`) on an 8-thread CPU; GPU and Apple silicon not measured |
+| gain over code alone in our benchmark | – | +5 points of doomed sessions caught | none measured (see below) |
+| data leaves the machine | no | yes (a scrubbed digest) | no |
+
+Choose with `UNWEDGE_PROVIDER` or the plugin option. Setup for each: [docs/providers.md](docs/providers.md).
+
+## Does it work?
+
+We replayed 218 public SWE-agent sessions (69 solved, 99 that failed after exhausting their
+context budget, 50 that submitted a wrong patch) through unwedge, with every threshold fixed
+before looking at the data. "Caught" means a doomed session got a hint or escalation before its
+last turn; a false alarm is the same thing in a session that went on to succeed.
+
+| policy (nothing tuned) | doomed sessions caught | successful sessions told they look stuck | successful sessions with any message | doomed-session spend after the first alarm |
+|---|---|---|---|---|
+| code only (`provider=none`) | 60% | 7.2% (5 of 69) | 7.2% | 47% |
+| **code + jev, as shipped** | **65%** | **7.2% (5 of 69)** | **15.9%** | **55%** |
+| jev only | 26% | 1.4% (1 of 69) | 10.1% | 27% |
+| original design: jev overrides code | 30% | 2.9% (2 of 69) | 11.6% | 31% |
+
+What we learned, plainly:
+
+- **Code does most of the work.** jev adds a modest gain on top: +5 points of catches and
+  +8 points of recoverable spend, with the same false-alarm rate. Its other contribution is a
+  one-time "the task may already be done; verify it and finish" note, which went to 9% of
+  successful sessions and to none of the failed ones. Letting the model *override* code halves
+  what is caught, so the shipped policy puts code first.
+- **No setting is precise enough to stop sessions automatically.** Even with thresholds tuned by
+  cross-validation, every detector (plain `max_turns` included) interrupted 1.4-5% of sessions
+  that would have succeeded. That is why hints are the default and stop mode is opt-in.
+- **jev reads windows well but predicts outcomes weakly.** Its stall judgment separates the
+  groups (median 0.70 in doomed sessions, 0.28 in successful ones), yet it rated most windows of
+  doomed sessions as still making some progress.
+- **Laya, as configured here, adds nothing yet.** On paired subsets of the same sessions, neither
+  checkpoint separated stuck from healthy sessions, so code + Laya performed exactly like code
+  alone. The integration works; its value for this task is not established.
+  [Details](docs/benchmark.md#laya).
+
+Method, cross-validated numbers, latency, cost, an edge-firewall finding and the caveats (one
+agent, one model family, 69 successes): [docs/benchmark.md](docs/benchmark.md). Everything is
+reproducible from [`benchmarks/`](benchmarks).
+
+## How it works
+
+Code signals on every turn → a provider only on suspicious turns (plus a periodic sample) → a
+pure-function policy with hysteresis, cooldowns and hint-before-stop → a hook response. The
+provider sees a small, secret-scrubbed digest of the session, sized to its context window.
+[docs/how-it-works.md](docs/how-it-works.md) · [docs/configuration.md](docs/configuration.md)
+
+## Privacy and security
+
+Tool output is scrubbed of common secrets and clipped before it is stored (`~/.unwedge`, deleted
+after 7 days) or sent to a provider. With `none` or `laya` nothing leaves your machine. unwedge is
+a cost and liveness guard, **not** a security control. See [SECURITY.md](SECURITY.md).
+
+## Status
+
+Alpha (0.1). The hook handler follows the documented Claude Code and Codex hook payloads and is
+covered by tests (Linux, macOS and Windows in CI). In a local Claude Code run the hooks fired
+and recorded the session; the rest of the flow was tested by feeding recorded hook payloads to
+`unwedge hook`. The Codex integration has not yet been run against a live Codex CLI. Transcript
+replay (`scan`, `replay`) reads internal formats and is best-effort. Thresholds were set before
+looking at the data and have not been tuned for Laya.
+
+## Contributing
+
+Issues and pull requests are welcome; see [CONTRIBUTING.md](CONTRIBUTING.md).
+
+## License and credits
+
+Apache-2.0. TypeSafe and jev are products of TypeSafe AI. Laya is by Convai Innovations;
+laya-mlx is an independent MLX port. The benchmark samples the public
+[nebius/SWE-agent-trajectories](https://huggingface.co/datasets/nebius/SWE-agent-trajectories)
+dataset, which is not redistributed here. This project is independent and not affiliated with
+any of them.