harnessie 0.6.0__tar.gz

harnessie-0.6.0/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 2026 Snap Synapse LLC
+
+   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.

harnessie-0.6.0/NOTICE

@@ -0,0 +1,38 @@
+Harnessie
+Copyright 2026 Snap Synapse LLC (https://snapsynapse.com/)
+
+Licensed under the Apache License, Version 2.0. See LICENSE.
+
+TRADEMARKS
+
+"Harnessie" and harnessie.com are marks of Snap Synapse LLC. Per Section 6
+of the Apache License, this license does not grant permission to use them.
+
+PAICE.WORK PBC STANDARDS AND SPECIFICATIONS
+
+Harnessie adopts lessons from, and credits, standards stewarded by
+PAICE.work PBC (https://paice.work/): Turnfile, AIDR (AI Decision Records),
+Graceful Boundaries, and GuideCheck. Those specifications, their names, and
+their canonical texts remain the property of PAICE.work PBC and are NOT
+licensed by this repository. Harnessie's relationship to them is adoption
+and credit, not ownership; conformance is asserted only where a tracked
+document cites the achieved level and its proving tests (see INTENT.md §7
+and GOVERNANCE.md). Nothing in this repository grants any right to the
+PAICE.work PBC specifications beyond what their own licenses provide.
+
+PAICE.WORK PBC ORIGINATED CODE
+
+As of this version, no PAICE.work PBC code is vendored into this
+repository. Standing rule: code originating in PAICE.work PBC repositories
+enters Harnessie only under an explicit written grant from PAICE.work PBC,
+recorded in this NOTICE with the grant date, the source, and the scope,
+before the vendored code appears in any public commit. Planned under the
+0.7.0 roadmap: a containment boundary adapted from PAICE.work PBC
+production PII-handling code; that adaptation will not merge before its
+grant line appears here.
+
+THIRD-PARTY MATERIALS
+
+This repository currently vendors no third-party code beyond the Python
+standard library. Future vendored components and their licenses will be
+credited here.

harnessie-0.6.0/PKG-INFO

@@ -0,0 +1,163 @@
+Metadata-Version: 2.4
+Name: harnessie
+Version: 0.6.0
+Summary: Harnessie: a brain-agnostic multi-agent harness (orchestrator / workers / verifiers) with verification gates, consent-based orchestration, ownership lanes, adversarial decision records, routing policy, and a tamper-evident audit log.
+Author-email: Sam Rogers <subscriptions@snapsynapse.com>
+License: Apache-2.0
+Project-URL: Homepage, https://harnessie.com/
+Project-URL: Repository, https://github.com/snapsynapse/harnessie
+Project-URL: Assistant-Guide, https://harnessie.com/.well-known/assistant-guide.txt
+Requires-Python: >=3.11
+Description-Content-Type: text/markdown
+License-File: LICENSE
+License-File: NOTICE
+Requires-Dist: pyyaml>=6.0
+Provides-Extra: dev
+Requires-Dist: pytest>=8.0; extra == "dev"
+Dynamic: license-file
+
+# Harnessie
+
+A brain-agnostic multi-agent harness: orchestrator, swappable workers, independent verifiers, verification gates between every phase, consent-based orchestration, per-agent file ownership, adversarial decision records with human-only arbitration, self-maintaining project memory with stamped provenance and dated expiry, cost routing as declared config, and a tamper-evident audit log that records operator and agent actions in one composite timeline.
+
+Harnessie is built to be the safest and easiest first AI harness for people. "Safest" is not a slogan here; it is a set of checkable properties: guarantees live in code, not prompts ([GOVERNANCE.md](GOVERNANCE.md)); every control that cannot be enforced fails closed instead of running unenforced; nothing ships on an agent's say-so (independent fresh-context verifiers gate every side-effecting phase); an OS sandbox confines shell work; a seven-layer prompt-injection defense with a written threat model ([SECURITY.md](SECURITY.md)) covers the rest; and a hash-chained audit log records every agent and operator action in one tamper-evident timeline. If you find a way to break any of these claims, that is a bug report we want. "Easiest" means the person running it does not need to be a developer: runs carry declared token and dollar ceilings, every halt is a named condition with one plain operator action, and disagreement between agents becomes a human decision, never a silent merge.
+
+The operating thesis: the harness structure carries the quality floor, the model carries the ceiling. Run it with Claude Fable 5 as the orchestrator and it exploits effort dials, long autonomous turns, and verifier subagents. Swap the workers (or everything) for Haiku, GLM, Qwen, or any OpenAI-compatible local endpoint by editing one YAML file, and the gates, jails, budgets, and retry ladders keep output honest.
+
+Beneath that thesis sit five engineering habits, each proven separately in the author's other tools and standards before it landed here as code: deterministic checks run before any model judgment (the gate order in [docs/GUIDE.md](docs/GUIDE.md)); evaluation comes before implementation — a governance mechanic without a red-then-green scenario pair does not merge ([EVALS.md](EVALS.md)); remembered facts carry verified and verify-by dates and expire visibly, never silently ([GOVERNANCE.md](GOVERNANCE.md)); every agent and operator action lands in one hash-chained, tamper-evident timeline ([docs/threat-model.md](docs/threat-model.md)); and any control that cannot be enforced fails closed rather than running unenforced ([SECURITY.md](SECURITY.md)). The same habits produced the standards Harnessie adopts — [Turnfile](https://turnfile.work/), [AIDR](https://aidr.work/), [Graceful Boundaries](https://gracefulboundaries.dev/) — which is why they fit together.
+
+## Quick start
+
+```bash
+pip install -e ".[dev]"
+python3 -m pytest -q                 # mock brain, no network
+python3 -m harness.cli eval          # deterministic eval scorecard
+python3 -m harness.cli verify-manifest
+export ANTHROPIC_API_KEY=sk-ant-...  # or point tiers at a local endpoint
+python3 -m harness.cli run workflows/build-and-verify.yaml --goal "a CLI todo app with tests"
+python3 -m harness.cli report <run_id>
+python3 -m harness.cli audit <run_id>   # verify the hash chain + governance timeline
+```
+
+Installed CLI usage can scaffold a fresh project layout:
+```bash
+harnessie init my-harnessie-project
+```
+
+Worked end-to-end example with sample data: [examples/policy-compliance/README.md](examples/policy-compliance/README.md).
+
+## Documentation
+
+- [docs/quickstart.md](docs/quickstart.md): the gentlest path for someone who has never cloned a repo or written YAML, assuming no git or shell fluency, with a glossary and an honest Windows/WSL2 page.
+- [docs/getting-started.md](docs/getting-started.md): the five-minute path from install to a green run and reading the evidence.
+- [docs/GUIDE.md](docs/GUIDE.md): the complete user guide, concepts through extension, including workflow authoring, brain configuration, ownership, governance, and the halt-recovery table.
+- [docs/brains.md](docs/brains.md): the brain-agnostic receipt, the models actually run under the harness with a link to the record that proves each.
+- [assistant-guide.txt](assistant-guide.txt): a bounded, human-verifiable guide (GuideCheck Level 3 profile) for an assistant reviewing a Harnessie checkout before you authorize a run; served at [harnessie.com/.well-known/assistant-guide.txt](https://harnessie.com/.well-known/assistant-guide.txt) with a sidecar manifest for provenance.
+
+The engineering references below (ARCHITECTURE, GOVERNANCE, SECURITY, ROADMAP) sit at the repo root; the user-facing guides live under `docs/`.
+
+## Layout
+
+```text
+harness/            the runtime: models/ tools/ loop verify routing memory state roles quarantine
+                    sandbox ownership adversarial audit runner cli events
+agents/             role prompts (markdown): orchestrator.md, workers/, verifiers/
+workflows/          declared phase sequences (YAML) with per-phase gates, task classes, and
+                    adversarial contested phases (mode: adversarial)
+config/models.yaml  model tiers, routing table, budgets: the ONLY file to edit to swap brains
+OWNERSHIP.yaml      ownership lanes + first-writer auto-claims; operator-owned, agents cannot reach it
+decisions/          the repo's own AIDR decision records (AIDR-0001 = v0.2 direction,
+                    AIDR-0002 = v0.3 direction; both human-arbitrated 2026-07-06
+                    with independent positions from four providers)
+memory/             project memory: MEMORY.md index + facts/ (stamped provenance, verify_by
+                    expiry) + archive/ (expired facts; nothing deletes) — maintained by
+                    workflows/memory-triage.yaml under approval gates
+examples/           worked end-to-end example (policy-compliance) with sample data
+runs/               per-run journal.jsonl (resume ledger) + events.jsonl (hash-chained audit)
+                    + proofs/ + decisions/ (contested-phase records) (gitignored)
+evals/              deterministic scorecards over mock-brain golden/risky/recovery scenarios
+docs/MANIFEST.yaml  trust-bundle integrity manifest for machine-readable public artifacts
+tests/              the done-tests for every subsystem
+docs/               reserved for the canonical web page (GitHub Pages publish source once public)
+*.md at root        ARCHITECTURE, GOVERNANCE, SECURITY, ROADMAP, IMPLEMENTATION_PLAN, PROMPTS, EVALS, NEXT,
+                    session-url-log, plus INTENT (9-section standard), CHANGELOG, README,
+                    LICENSE (Apache-2.0) + NOTICE; source-verification.json is the build provenance data
+```
+
+Dogfooding this repo under Claude Code uses a local `.claude/` (subagent defs, a `/run-workflow` command, a pytest hook). Per repo convention `.claude/` is gitignored, so it does not ship; the canonical role prompts it wraps live in `agents/`, and the CLI is the primary interface.
+
+## Requirements
+
+Python 3.11+ and PyYAML (installed by `pip install -e .`). The stdlib-only model adapters need no vendor SDK. The OS sandbox uses native macOS `sandbox-exec` when it can actually apply a Seatbelt profile; on Linux it uses bubblewrap, firejail, or docker (in that order of preference). Every backend is admitted only after a startup smoke test proves it can confine here; a present-but-unusable backend, and any platform with none (Windows), fails closed so shell-using workflows are blocked rather than run unconfined (see [SECURITY.md](SECURITY.md)).
+
+Live provider scorecards are opt-in and never part of the default no-network suite. With credentials and a local endpoint configured, run:
+```bash
+HARNESSIE_LIVE=1 \
+HARNESSIE_OPENAI_COMPAT_BASE_URL=http://localhost:11434/v1 \
+python3 -m harness.cli eval --live
+```
+Without `HARNESSIE_LIVE=1` or provider configuration, the live scorecard reports explicit `SKIP` rows and exits cleanly.
+
+## Design in one breath
+
+Goal enters, orchestrator (frontier, high effort) decomposes it into task packets with acceptance criteria and out-of-scope fences; task packets are offers, and workers consent (or decline with a counter-proposal) before side-effecting tools unlock; workers (cheap tiers) execute inside a jailed workspace with allowlisted tools, owning the files they create and never each other's; every worker phase exits through a gate that runs deterministic checks first, then an independent fresh-context verifier that never sees the worker's reasoning and fails closed; contested decisions fan out to an adversarial panel whose positions, objections, and dissent land in a decision record only a human may arbitrate; failures reformulate the task with evidence and escalate effort-then-tier before halting for a human; everything is journaled, budgeted, resumable, hash-chain audited, and leaves proof artifacts on disk.
+
+For long runs, approval-gated tools can be authorized by a small headless policy file:
+```yaml
+allow:
+  - tool: expire_fact
+    phase: triage
+deny:
+  - tool: deploy
+```
+Run with `--approval-policy approvals.yaml`, or use `--approve-interactive` to prompt on a TTY. Independent phases can fan out by sharing a `parallel:` label; each runs under `workspace/.phases/<phase>` and gates independently before later phases see its report.
+
+Optional extra review can use local OpenAI-compatible endpoints such as Ollama or CLI fan-out across agents. That is useful evidence, especially for patches touching orchestration, but it does not replace the deterministic suite: `pytest`, `harnessie eval`, manifest verification, and scrub/audit checks remain the proof surface.
+
+Full rationale and the verified source-to-decision map: [ARCHITECTURE.md](ARCHITECTURE.md). Governance layer (consent, ownership, contest, audit): [GOVERNANCE.md](GOVERNANCE.md). Prompt-injection and secret-handling model: [SECURITY.md](SECURITY.md). The "safest" claim as a falsifiable table, each row citing enforcing code and its test, versus prevailing harness patterns: [docs/threat-model.md](docs/threat-model.md). What comes next and platform support: [ROADMAP.md](ROADMAP.md).
+
+## What governs a run
+
+A run's behavior is not in one file; each decision has one owner. To predict or change what a run will do, edit the owner, not the prompt:
+
+| Decision | Governed by |
+|---|---|
+| Which model runs each task class, and how to swap brains | `config/models.yaml` (tiers + routing table) |
+| Token and dollar ceilings, effort per task class | `config/models.yaml` (budget + routing) |
+| Which phases run, in what order, with which gates and verifiers | the workflow YAML in `workflows/` |
+| Which files each agent may write | `OWNERSHIP.yaml` (lanes + first-writer claims) |
+| What each role may do (tools, shell allowlist, approval) | the tool registry (`harness/tools/builtin.py`) + role prompts in `agents/` |
+
+## When a run halts
+
+Silence is never success: every run ends in a named stop condition, and each maps to one operator action. Resuming is `harnessie run <same workflow> --goal ...` with the same run id: resume re-runs only phases that did not pass, so fixing the cause and re-running is safe.
+
+| Stop condition | What it means | What to do |
+|---|---|---|
+| `complete` / phase `passed` | task done, gate satisfied | nothing; the next phase proceeds |
+| `declined` | the worker declined the offered task packet | read the counter-proposal in the report; revise the packet or accept the objection, then re-run |
+| `needs_human` | a gate's checks or verifier failed after the retry ladder exhausted | read the proof artifacts under `runs/<id>/proofs/` and the report; fix the task or the acceptance criteria; re-run |
+| `needs_arbitration` | a contested phase produced dissent | open `runs/<id>/decisions/DR-<phase>.md`, record your arbitration decision in it, then re-run (resume keys on that record) |
+| `stuck` | the model repeated an identical failing or refused call | inspect the refusal (`harnessie audit <id>`); fix the tool grant, allowlist, or task; re-run |
+| `budget` | the run hit its token or dollar ceiling | raise the ceiling in `config/models.yaml` or narrow the goal; re-run |
+| `max_steps` | the loop hit its step ceiling without completing | raise `max_steps` for the phase or simplify the task |
+| `model_error` | the provider errored twice in a row | check the endpoint and API key; re-run |
+| `no_action` | the model produced no tool call even after a nudge | usually a role-prompt or model-fit issue; check the role prompt in `agents/` |
+
+## Built on open standards
+
+Harnessie's governance mechanics are code-enforced imports of two open, vendor-neutral standards, and the design philosophy beneath both:
+
+- [Turnfile](https://turnfile.work/): consent-based coordination, ownership lanes, authority order, and bounded rebuttal became the task-packet offer contract, `OWNERSHIP.yaml`, and the objection rounds in contested phases.
+- [AIDR](https://aidr.work/): the decision-record lifecycle, preserved dissent, human-only arbitration, and structurally earned claims became the contested-phase records in `decisions/` and `runs/<id>/decisions/`. This repo dogfoods AIDR for its own direction decisions.
+- [The Aggregated Intelligence tenets](https://paice.foundation/papers/aggregated-intelligence-tenets.html): intelligence lives in the arrangement, not the node; disagreement is the engine, not the exhaust; independence before influence; consensus is evidence, never authority; authority is human because accountability is human. The full tenet-to-mechanism mapping is [GOVERNANCE.md](GOVERNANCE.md) §7.
+
+These are lesson imports, not conformance claims: Harnessie asserts no Turnfile or AIDR conformance. If the harness makes you ask why its rules work, those standards are the answer.
+
+## Contributing
+
+See [CONTRIBUTING.md](CONTRIBUTING.md) for setup, the eval-first change discipline, and how consequential decisions are recorded. Bug and feature templates live under [.github/](.github/).
+
+## License
+
+Apache-2.0. See [LICENSE](LICENSE) and [NOTICE](NOTICE) (copyright Snap Synapse LLC; trademark and PAICE.work PBC spec/code carveouts).

harnessie-0.6.0/README.md

@@ -0,0 +1,145 @@
+# Harnessie
+
+A brain-agnostic multi-agent harness: orchestrator, swappable workers, independent verifiers, verification gates between every phase, consent-based orchestration, per-agent file ownership, adversarial decision records with human-only arbitration, self-maintaining project memory with stamped provenance and dated expiry, cost routing as declared config, and a tamper-evident audit log that records operator and agent actions in one composite timeline.
+
+Harnessie is built to be the safest and easiest first AI harness for people. "Safest" is not a slogan here; it is a set of checkable properties: guarantees live in code, not prompts ([GOVERNANCE.md](GOVERNANCE.md)); every control that cannot be enforced fails closed instead of running unenforced; nothing ships on an agent's say-so (independent fresh-context verifiers gate every side-effecting phase); an OS sandbox confines shell work; a seven-layer prompt-injection defense with a written threat model ([SECURITY.md](SECURITY.md)) covers the rest; and a hash-chained audit log records every agent and operator action in one tamper-evident timeline. If you find a way to break any of these claims, that is a bug report we want. "Easiest" means the person running it does not need to be a developer: runs carry declared token and dollar ceilings, every halt is a named condition with one plain operator action, and disagreement between agents becomes a human decision, never a silent merge.
+
+The operating thesis: the harness structure carries the quality floor, the model carries the ceiling. Run it with Claude Fable 5 as the orchestrator and it exploits effort dials, long autonomous turns, and verifier subagents. Swap the workers (or everything) for Haiku, GLM, Qwen, or any OpenAI-compatible local endpoint by editing one YAML file, and the gates, jails, budgets, and retry ladders keep output honest.
+
+Beneath that thesis sit five engineering habits, each proven separately in the author's other tools and standards before it landed here as code: deterministic checks run before any model judgment (the gate order in [docs/GUIDE.md](docs/GUIDE.md)); evaluation comes before implementation — a governance mechanic without a red-then-green scenario pair does not merge ([EVALS.md](EVALS.md)); remembered facts carry verified and verify-by dates and expire visibly, never silently ([GOVERNANCE.md](GOVERNANCE.md)); every agent and operator action lands in one hash-chained, tamper-evident timeline ([docs/threat-model.md](docs/threat-model.md)); and any control that cannot be enforced fails closed rather than running unenforced ([SECURITY.md](SECURITY.md)). The same habits produced the standards Harnessie adopts — [Turnfile](https://turnfile.work/), [AIDR](https://aidr.work/), [Graceful Boundaries](https://gracefulboundaries.dev/) — which is why they fit together.
+
+## Quick start
+
+```bash
+pip install -e ".[dev]"
+python3 -m pytest -q                 # mock brain, no network
+python3 -m harness.cli eval          # deterministic eval scorecard
+python3 -m harness.cli verify-manifest
+export ANTHROPIC_API_KEY=sk-ant-...  # or point tiers at a local endpoint
+python3 -m harness.cli run workflows/build-and-verify.yaml --goal "a CLI todo app with tests"
+python3 -m harness.cli report <run_id>
+python3 -m harness.cli audit <run_id>   # verify the hash chain + governance timeline
+```
+
+Installed CLI usage can scaffold a fresh project layout:
+```bash
+harnessie init my-harnessie-project
+```
+
+Worked end-to-end example with sample data: [examples/policy-compliance/README.md](examples/policy-compliance/README.md).
+
+## Documentation
+
+- [docs/quickstart.md](docs/quickstart.md): the gentlest path for someone who has never cloned a repo or written YAML, assuming no git or shell fluency, with a glossary and an honest Windows/WSL2 page.
+- [docs/getting-started.md](docs/getting-started.md): the five-minute path from install to a green run and reading the evidence.
+- [docs/GUIDE.md](docs/GUIDE.md): the complete user guide, concepts through extension, including workflow authoring, brain configuration, ownership, governance, and the halt-recovery table.
+- [docs/brains.md](docs/brains.md): the brain-agnostic receipt, the models actually run under the harness with a link to the record that proves each.
+- [assistant-guide.txt](assistant-guide.txt): a bounded, human-verifiable guide (GuideCheck Level 3 profile) for an assistant reviewing a Harnessie checkout before you authorize a run; served at [harnessie.com/.well-known/assistant-guide.txt](https://harnessie.com/.well-known/assistant-guide.txt) with a sidecar manifest for provenance.
+
+The engineering references below (ARCHITECTURE, GOVERNANCE, SECURITY, ROADMAP) sit at the repo root; the user-facing guides live under `docs/`.
+
+## Layout
+
+```text
+harness/            the runtime: models/ tools/ loop verify routing memory state roles quarantine
+                    sandbox ownership adversarial audit runner cli events
+agents/             role prompts (markdown): orchestrator.md, workers/, verifiers/
+workflows/          declared phase sequences (YAML) with per-phase gates, task classes, and
+                    adversarial contested phases (mode: adversarial)
+config/models.yaml  model tiers, routing table, budgets: the ONLY file to edit to swap brains
+OWNERSHIP.yaml      ownership lanes + first-writer auto-claims; operator-owned, agents cannot reach it
+decisions/          the repo's own AIDR decision records (AIDR-0001 = v0.2 direction,
+                    AIDR-0002 = v0.3 direction; both human-arbitrated 2026-07-06
+                    with independent positions from four providers)
+memory/             project memory: MEMORY.md index + facts/ (stamped provenance, verify_by
+                    expiry) + archive/ (expired facts; nothing deletes) — maintained by
+                    workflows/memory-triage.yaml under approval gates
+examples/           worked end-to-end example (policy-compliance) with sample data
+runs/               per-run journal.jsonl (resume ledger) + events.jsonl (hash-chained audit)
+                    + proofs/ + decisions/ (contested-phase records) (gitignored)
+evals/              deterministic scorecards over mock-brain golden/risky/recovery scenarios
+docs/MANIFEST.yaml  trust-bundle integrity manifest for machine-readable public artifacts
+tests/              the done-tests for every subsystem
+docs/               reserved for the canonical web page (GitHub Pages publish source once public)
+*.md at root        ARCHITECTURE, GOVERNANCE, SECURITY, ROADMAP, IMPLEMENTATION_PLAN, PROMPTS, EVALS, NEXT,
+                    session-url-log, plus INTENT (9-section standard), CHANGELOG, README,
+                    LICENSE (Apache-2.0) + NOTICE; source-verification.json is the build provenance data
+```
+
+Dogfooding this repo under Claude Code uses a local `.claude/` (subagent defs, a `/run-workflow` command, a pytest hook). Per repo convention `.claude/` is gitignored, so it does not ship; the canonical role prompts it wraps live in `agents/`, and the CLI is the primary interface.
+
+## Requirements
+
+Python 3.11+ and PyYAML (installed by `pip install -e .`). The stdlib-only model adapters need no vendor SDK. The OS sandbox uses native macOS `sandbox-exec` when it can actually apply a Seatbelt profile; on Linux it uses bubblewrap, firejail, or docker (in that order of preference). Every backend is admitted only after a startup smoke test proves it can confine here; a present-but-unusable backend, and any platform with none (Windows), fails closed so shell-using workflows are blocked rather than run unconfined (see [SECURITY.md](SECURITY.md)).
+
+Live provider scorecards are opt-in and never part of the default no-network suite. With credentials and a local endpoint configured, run:
+```bash
+HARNESSIE_LIVE=1 \
+HARNESSIE_OPENAI_COMPAT_BASE_URL=http://localhost:11434/v1 \
+python3 -m harness.cli eval --live
+```
+Without `HARNESSIE_LIVE=1` or provider configuration, the live scorecard reports explicit `SKIP` rows and exits cleanly.
+
+## Design in one breath
+
+Goal enters, orchestrator (frontier, high effort) decomposes it into task packets with acceptance criteria and out-of-scope fences; task packets are offers, and workers consent (or decline with a counter-proposal) before side-effecting tools unlock; workers (cheap tiers) execute inside a jailed workspace with allowlisted tools, owning the files they create and never each other's; every worker phase exits through a gate that runs deterministic checks first, then an independent fresh-context verifier that never sees the worker's reasoning and fails closed; contested decisions fan out to an adversarial panel whose positions, objections, and dissent land in a decision record only a human may arbitrate; failures reformulate the task with evidence and escalate effort-then-tier before halting for a human; everything is journaled, budgeted, resumable, hash-chain audited, and leaves proof artifacts on disk.
+
+For long runs, approval-gated tools can be authorized by a small headless policy file:
+```yaml
+allow:
+  - tool: expire_fact
+    phase: triage
+deny:
+  - tool: deploy
+```
+Run with `--approval-policy approvals.yaml`, or use `--approve-interactive` to prompt on a TTY. Independent phases can fan out by sharing a `parallel:` label; each runs under `workspace/.phases/<phase>` and gates independently before later phases see its report.
+
+Optional extra review can use local OpenAI-compatible endpoints such as Ollama or CLI fan-out across agents. That is useful evidence, especially for patches touching orchestration, but it does not replace the deterministic suite: `pytest`, `harnessie eval`, manifest verification, and scrub/audit checks remain the proof surface.
+
+Full rationale and the verified source-to-decision map: [ARCHITECTURE.md](ARCHITECTURE.md). Governance layer (consent, ownership, contest, audit): [GOVERNANCE.md](GOVERNANCE.md). Prompt-injection and secret-handling model: [SECURITY.md](SECURITY.md). The "safest" claim as a falsifiable table, each row citing enforcing code and its test, versus prevailing harness patterns: [docs/threat-model.md](docs/threat-model.md). What comes next and platform support: [ROADMAP.md](ROADMAP.md).
+
+## What governs a run
+
+A run's behavior is not in one file; each decision has one owner. To predict or change what a run will do, edit the owner, not the prompt:
+
+| Decision | Governed by |
+|---|---|
+| Which model runs each task class, and how to swap brains | `config/models.yaml` (tiers + routing table) |
+| Token and dollar ceilings, effort per task class | `config/models.yaml` (budget + routing) |
+| Which phases run, in what order, with which gates and verifiers | the workflow YAML in `workflows/` |
+| Which files each agent may write | `OWNERSHIP.yaml` (lanes + first-writer claims) |
+| What each role may do (tools, shell allowlist, approval) | the tool registry (`harness/tools/builtin.py`) + role prompts in `agents/` |
+
+## When a run halts
+
+Silence is never success: every run ends in a named stop condition, and each maps to one operator action. Resuming is `harnessie run <same workflow> --goal ...` with the same run id: resume re-runs only phases that did not pass, so fixing the cause and re-running is safe.
+
+| Stop condition | What it means | What to do |
+|---|---|---|
+| `complete` / phase `passed` | task done, gate satisfied | nothing; the next phase proceeds |
+| `declined` | the worker declined the offered task packet | read the counter-proposal in the report; revise the packet or accept the objection, then re-run |
+| `needs_human` | a gate's checks or verifier failed after the retry ladder exhausted | read the proof artifacts under `runs/<id>/proofs/` and the report; fix the task or the acceptance criteria; re-run |
+| `needs_arbitration` | a contested phase produced dissent | open `runs/<id>/decisions/DR-<phase>.md`, record your arbitration decision in it, then re-run (resume keys on that record) |
+| `stuck` | the model repeated an identical failing or refused call | inspect the refusal (`harnessie audit <id>`); fix the tool grant, allowlist, or task; re-run |
+| `budget` | the run hit its token or dollar ceiling | raise the ceiling in `config/models.yaml` or narrow the goal; re-run |
+| `max_steps` | the loop hit its step ceiling without completing | raise `max_steps` for the phase or simplify the task |
+| `model_error` | the provider errored twice in a row | check the endpoint and API key; re-run |
+| `no_action` | the model produced no tool call even after a nudge | usually a role-prompt or model-fit issue; check the role prompt in `agents/` |
+
+## Built on open standards
+
+Harnessie's governance mechanics are code-enforced imports of two open, vendor-neutral standards, and the design philosophy beneath both:
+
+- [Turnfile](https://turnfile.work/): consent-based coordination, ownership lanes, authority order, and bounded rebuttal became the task-packet offer contract, `OWNERSHIP.yaml`, and the objection rounds in contested phases.
+- [AIDR](https://aidr.work/): the decision-record lifecycle, preserved dissent, human-only arbitration, and structurally earned claims became the contested-phase records in `decisions/` and `runs/<id>/decisions/`. This repo dogfoods AIDR for its own direction decisions.
+- [The Aggregated Intelligence tenets](https://paice.foundation/papers/aggregated-intelligence-tenets.html): intelligence lives in the arrangement, not the node; disagreement is the engine, not the exhaust; independence before influence; consensus is evidence, never authority; authority is human because accountability is human. The full tenet-to-mechanism mapping is [GOVERNANCE.md](GOVERNANCE.md) §7.
+
+These are lesson imports, not conformance claims: Harnessie asserts no Turnfile or AIDR conformance. If the harness makes you ask why its rules work, those standards are the answer.
+
+## Contributing
+
+See [CONTRIBUTING.md](CONTRIBUTING.md) for setup, the eval-first change discipline, and how consequential decisions are recorded. Bug and feature templates live under [.github/](.github/).
+
+## License
+
+Apache-2.0. See [LICENSE](LICENSE) and [NOTICE](NOTICE) (copyright Snap Synapse LLC; trademark and PAICE.work PBC spec/code carveouts).

harnessie-0.6.0/harness/__init__.py

@@ -0,0 +1,25 @@
+"""Harnessie, a brain-agnostic multi-agent harness.
+
+Subsystem map (each module is one explicit boundary):
+
+    models/       ModelInterface + provider adapters (Anthropic, OpenAI-compatible, mock)
+    tools/        ToolRegistry: single source of truth for capabilities AND policy
+    loop.py       AgentLoop: goal -> context -> model -> permission gate -> tool -> repeat
+    verify.py     VerificationGate: deterministic checks + model verifier + retry ladder
+    routing.py    Routing policy: task tier -> (model, effort), escalation ladder, budgets
+    memory.py     Project memory (markdown facts w/ provenance) + proof artifacts
+    state.py      Run journal (JSONL) for resumability and audit
+    roles.py      Role definitions (orchestrator / worker / verifier) loaded from agents/
+    quarantine.py Prompt-injection ingress filter + secret detection/redaction
+    sandbox.py    OS confinement of child commands (workspace-only writes, network deny)
+    ownership.py  Ownership lanes: agents own their files, never each other's
+    adversarial.py Contested phases: positions, objections, decision records, lint
+    audit.py      Hash-chain verification + governance timeline for any run
+    runner.py     WorkflowRunner: executes workflows/*.yaml phase by phase through gates
+    events.py     Structured, hash-chained event log shared by everything above
+
+Security model: SECURITY.md. Architecture and source map: ARCHITECTURE.md.
+Governance layer (consent, ownership, contest, audit): GOVERNANCE.md.
+"""
+
+__version__ = "0.5.0"