solaris-dev-shop-native 0.0.0-stage → 0.66.0

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Solaris Dev Shop
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/PI-DISTRIBUTION.md

@@ -0,0 +1,11 @@
+# Solaris Dev Shop for Pi 0.66.0
+
+61 specialists and one coordinator; 1,200 canonical files and 1,979 complete contained files/modes. Install the complete package from the reviewed release:
+
+```sh
+pi install git:github.com/Shaisolaris/solaris-dev-shop-plugins@v0.66.0
+```
+
+The root manifest declares 62 exact skill paths and no extensions, prompts, themes, dependencies or lifecycle scripts. Source checks exercise all contexts, coordinator and SQLite recovery. A current native client result must be bound to the actual installed revision; [older results](docs/history/HARNESS-DISTRIBUTION-064.md) remain historical.
+
+npm publication and gallery discovery require an actual authenticated registry publication; Git availability does not establish that outcome. Python and individual role prerequisites apply. No Docker startup, service setup, self-update or learning rewrite occurs on workflow loading. See [validation](VALIDATION.md).

package/README.md

@@ -1,3 +1,64 @@
-# Temporary Holding Version
+# Solaris Dev Shop native plugin
 
-This version is a temporary placeholder for this package. An operational version to replace this has been submitted for review and is awaiting a staged release.
+61 specialist workflows, a chief of staff and executable local helpers. Published by
+Solaris Technologies. Generated from [Solaris Dev Shop 0.66](https://github.com/Shaisolaris/solaris-dev-shop/tree/cd1898c6dc062a2dcf01a56b53403023c1994c8d).
+
+## Install
+
+### Codex
+
+```sh
+codex plugin marketplace add Shaisolaris/solaris-dev-shop-plugins
+codex plugin add solaris-dev-shop@solaris-technologies
+```
+
+### Claude Code
+
+```text
+/plugin install solaris-dev-shop --marketplace Shaisolaris/solaris-dev-shop-plugins
+```
+
+Or add this repository as a marketplace and install `solaris-dev-shop@solaris-technologies`.
+
+### OpenCode V2
+
+```sh
+opencode plugin add 'github:Shaisolaris/solaris-dev-shop-plugins#main::path:plugin'
+```
+
+For a pinned installation, replace `main` with the complete distribution commit.
+For a local installation, add the extracted `plugin` directory to `plugins` in
+`opencode.json`. This adapter uses the V2 API, tested with CLI/SDK 2.1.2.
+
+## Use
+
+Ask the host to use the Solaris chief of staff or a named specialist. The native
+skill entry loads the canonical workflow and its companion files. Run helpers
+from their canonical role folders. See [the capability guide](plugin/docs/capabilities.md)
+for runnable examples, and [the unchanged upstream README](plugin/SOURCE-README.md).
+
+The source package contains specialist resources. The optional Python wheel
+contains the deterministic router and catalog rather than every helper.
+
+## Scope and evidence
+
+The plugin adds no account access, standing tool permissions, telemetry service,
+MCP connection or background schedule. The host supplies tools and permissions.
+Local helpers retain their documented runtime and operating-system limits.
+
+The 61-role roster is not evidence that every role gained new executable behavior:
+the 0.62 programme records changed-role source/runtime evidence for 37 roles.
+Examples are bounded local or synthetic workflows, not production certification.
+Host discovery and installed-helper execution are checked separately from model
+activation. Native format support is separate from curated directory approval.
+
+OpenCode's model/provider configuration is independent of the plugin. No
+standalone DeepSeek plugin marketplace approval is claimed.
+
+## Support
+
+Use [GitHub issues](https://github.com/Shaisolaris/solaris-dev-shop-plugins/issues) or email shehryar@solaristek.com.
+Do not post credentials, personal data or private source code in public issues.
+The software is distributed under the upstream [MIT license](plugin/LICENSE).
+See [plugin data handling](plugin/DATA-HANDLING.md) for local files, host/provider
+processing and support contacts. The publisher website policy has its own stated scope.

package/copilot/plugin/.claude-plugin/plugin.json

@@ -0,0 +1,14 @@
+{
+  "name": "solaris-dev-shop",
+  "version": "0.66.0",
+  "description": "61 specialist workflows, a chief of staff and executable local helpers.",
+  "author": {
+    "name": "Solaris Technologies",
+    "email": "shehryar@solaristek.com",
+    "url": "https://solaristek.com"
+  },
+  "homepage": "https://github.com/Shaisolaris/solaris-dev-shop-plugins",
+  "repository": "https://github.com/Shaisolaris/solaris-dev-shop-plugins",
+  "license": "MIT",
+  "icon": "./assets/native-plugin-icon.png"
+}

package/copilot/plugin/CHANGELOG.md

@@ -0,0 +1,146 @@
+# Changelog
+
+## 0.66 (2026-10-23)
+
+- Make runtime setup explicit throughout the complete 61-specialist product. Loading a workflow does not authorize Docker, MCP/database services, background jobs or source refresh. Keep references on demand rather than installing an eager all-skill startup hook.
+- Remove the blanket Docker requirement and unverified agent-box installer. Container examples use reviewed image digests, refuse mutable images, prevent implicit pulls and scope mounts, network and lifetime to the task. Containers and workflow flags are not presented as OS permission boundaries.
+- Require existing container execution scope before WordPress database commands; remove the automatic CLI-container fallback and host-network mode. Keep local-only database publication, cleanup and literal path arguments.
+- Leave CAD and Unreal server auto-start disabled. Make worker provisioning an optional operator procedure and replace the remote mobile-tool shell-pipe installer with reviewed setup.
+- Install only from a reviewed local checkout, without cloning, silently updating it or running developer tests that fetch build dependencies. The installer runs bounded offline routing checks. Preserve existing files on refusal and retain the complete roster, original branding and prior immutable releases. Native format compatibility and provider marketplace approval remain separate.
+- Keep mobile/editor remote automation on task-scoped, authenticated loopback connections; remove wildcard unauthenticated listener defaults, unpinned MCP startup downloads and automatic fallback setup. Missing WordPress tooling yields a reviewed setup prerequisite instead of an unsigned privileged installation recipe.
+- Disable automatic learning promotions. Pending source promotions now produce a real review request instead of falling through to assignment; ordinary work remains assignable. Keep new observations in project notes and installed workflow files unchanged.
+
+## 0.65 (2026-10-23)
+
+- Refuse WordPress plugin installation when its API response is missing, malformed, mismatched or unsuccessful. Validate plugin names before requests, encode API parameters and keep response JSON in a bounded, typed reader isolated from project modules.
+- Pass WordPress modification paths, requests and JSON values as data to literal Python programs. Preserve existing build state when history or replacement JSON is invalid.
+- Require explicitly selected repository permissions and short-lived worker credentials. Remove the account-wide OAuth fallback, keep login and disk-encryption protection, and distinguish policy declarations from actual operating-system and credential enforcement.
+- Change Unreal project standing instructions only on the owner's explicit request. Preserve the complete 61-specialist roster, coordinator and original branding.
+- Integrate these reviewed security corrections as one distribution milestone. Native loading tests, public source publication and provider marketplace approval remain separate outcomes.
+
+## 0.64 (2026-10-22)
+
+- Preserve an existing skill-copy destination instead of deleting its files. Refuse files, directories, broken symlinks and destinations inside the source checkout; fresh copies and recovery remain usable.
+- Keep upstream TLS verification enabled in all network-role instructions. Private devices use an explicitly trusted CA and expected certificate name, with live trusted/untrusted/name-mismatch regression checks.
+- Treat WordPress build paths and flags as data rather than Python source. Quote configuration directories for the remote shell; executable instruction tests cover ordinary output, metacharacter refusal and recovery without contacting a real site.
+- Bind the security corrections to exact source components and the declared integration scope. Preserve the 61 specialists and coordinator, original branding, useful local helpers and documented platform limits.
+- Carry the corrected source into the native distribution programme. Host compatibility and directory approval remain separate; the native repository records each tested client and exact install revision.
+
+## 0.63 (2026-10-22)
+
+- Install the same specialist workflows through the native Codex/OpenAI, Claude Code and OpenCode V2 repository distribution. The original 61 roles and coordinator remain intact; installation and discovery are separate from model performance and provider-directory approval.
+- Keep offline invoice intent unique when clients enqueue or resolve the same change concurrently. Retry creation and supersession now commit together, and malformed legacy outboxes stop before any request is sent.
+- Keep synthetic loopback credentials on their assigned route: offline sync refuses redirects and ignores environment proxy settings while preserving pending changes.
+- Return bounded client errors for malformed Unicode or oversized numeric API fields instead of dropping the connection. Ordinary decimal values, including leading zeroes, retain their behavior.
+- Bind generated native loaders, adapter instructions, lifecycle metadata and complete marketplace catalogs to an independently pinned packager. Both built-in controls and a separate independent tampering review check the generated surface.
+
+## 0.62 (2026-10-21)
+
+### Executable specialist workflows
+
+- Resume interrupted work orders without sending the same delivery twice. Local activation preserves owner material, and fleet registration tracks claims and recovery state.
+- Rehearse SQLite backup/restore and approved local deployment rollback. Diagnose saved Kubernetes, network, incident, WordPress and page evidence with explicit uncertainty and refusal states.
+- Reconcile dirty CSV exports, invoices, payments and cash by currency before using their figures. Experiment readouts distinguish an observed difference from permission to make a causal claim; missing population evidence and cross-arm exclusions block that claim.
+- Evaluate grounded answers against local source snapshots, preserve citation offsets and abstain on recognized instruction-like source context. The filter has documented lexical limits and conservative exclusions; it is not a general prompt-injection defense.
+- Exercise duplicate-safe API requests, offline synchronization, signed webhook ledgers, inventory conservation and recorded exception resolution using local synthetic fixtures.
+- Compare product options against pinned quotes, preserve percentage units and reject numeric fragments. Check scope acceptance, consent-aware follow-up drafts, privacy inventories and license/provenance consistency without treating them as permission to send or a legal determination.
+- Add a stale-response UI demonstration, replayable device and event-chain simulations, and inspectable campaign accounting. Run a complete cross-employee delivery drill with forced interruption, another-worker resume and invalidated stale evidence.
+
+### Quality and installation
+
+- Start the loopback API and its negative controls without a reverse DNS lookup, avoiding offline macOS startup hangs.
+
+- Bind delivery evidence to the command, tool and artifact bytes; expose rewritten inputs as gaps and reject stale outputs. Release inventory requires candidate, test and review assertions to name the same complete revision. Repeated input/output flags retain every path. Integration checks reuse ancestry answers within one invocation and refresh them on the next.
+- Keep the original 61 specialist roles and branding. The wheel still packages the deterministic router; specialist workflows and companion resources are distributed in the source tree.
+- Test full capability regressions on supported Linux/macOS runtimes and the router, assurance, platform boundaries and clean wheel on Windows. Portable skill text does not imply every helper supports every operating system.
+
+### Compatibility
+
+- Grounded-evaluation journals now bind the evaluator source hash under journal schema 3. Keep an old journal for audit and start a new one after an engine change; older journals are refused rather than silently resumed.
+- All delivery, provider, infrastructure, financial and communication examples remain bounded by their documented environments and host permissions. No new account access, live deployment, real payment, message sending or provider integration is granted.
+
+## 0.61 (2026-10-18)
+
+- Harden assurance finding closure and human waivers: reject expired or malformed waivers, malformed severity/status, blank closers and invalid evidence containers before mutation.
+- Preserve legitimate independent closure and valid human-waiver closure, including empty evidence where a valid waiver authorizes it. Authority and evidence remain declared inputs, not authenticated or dereferenced proof.
+- Add 40 focused API and real CLI regression tests; independently verify supported Python 3.12, fixed-clock expiry boundaries and correction/retry behavior.
+
+## 0.60 (2026-10-16)
+
+- Enforce typed, nonblank Quality OS gate evidence and block receipts reporting open findings. Malformed packet containers now yield diagnostics instead of false passes or tracebacks.
+- Reject unknown gate/dimension inputs and nested validation failures; document receipt validation versus actual execution and owner approval.
+- Add independent regression coverage for valid and malformed handoffs. Assurance engine is 1.0.1; intake engine/schema remain 1.3.0 and the roster remains 61.
+
+## 0.59 (2026-10-15)
+
+- Package the canonical router directly without source symlinks or duplicate package layout, including source-to-wheel validation.
+
+## 0.58 (2026-10-15)
+
+- Preserve requested supporting specialists and source handoffs in compound assignments; add contributor quickstart and corrected new-skill instructions.
+
+## 0.57 (2026-10-15)
+
+- Repair 14 malformed YAML descriptions and 22 discovery-metadata compatibility failures in total, including overlong fields and bracket comparisons. Skill names and detailed bodies remain unchanged.
+- Add a safe YAML/name/description gate for all 61 specialists plus the coordinator. PyYAML is a developer-only dependency; the intake runtime remains standard-library only.
+- Retain engine 1.2.1 and the existing roster. No provider, tool permission or specialist is added.
+- Restore the original project preview artwork and preserve its branding; custom preview settings are unchanged.
+
+## 0.56 (2026-10-15)
+
+- Recognize production action and target words even when an application name separates them; run the human-review risk gate before ambiguity or context routing.
+- Cover ordinary deployment/release wording and non-production/whole-word boundaries with regressions. Document conservative lexical matching and host permission limits.
+
+## 0.55 (2026-10-15)
+
+- Ask for clarification on long unknown requests instead of silently assigning a backend specialist. Retain specific keyword and multi-domain routes.
+- Add offline skill search by terms and department, plus source paths and links in assignment packets. Validate the catalog against all 61 active skills in CI.
+- Align the 3D Artist contract ID with its existing roster and route. No roster additions.
+- Provide a complete clone-to-handoff quickstart, a worked first-task guide and verifiable project facts. Clarify host execution and benchmark limits.
+- Repair 34 shared-standard paths so they resolve from the skill folder, with a source-file regression check. Document shared dependencies when copying skills.
+- Refresh repository preview assets with matching editable source and 2:1 output.
+
+## 0.54 (2026-10-15)
+
+- Fixed self-test fixture discovery: checkout fixtures are now executed and their expectations enforced.
+- Added `intake --explain`, relative keyword confidence and distinct runner-up evidence.
+- Added bounded Spanish/French review phrases, matching risk signals and Unicode word boundaries.
+- Added opt-in local follow-up routing context with no stored request text; safety gates run on every request.
+- Added escaped, offline Quality OS HTML reports.
+- Expanded adversarial fixtures from 3 to 14 and added CLI, privacy, Unicode and installer regressions.
+- Added actual multi-domain/escalation examples and a reproducible 100-request benchmark.
+- Added a read-only Actions routing playground and Windows manual installation checks.
+- CI now checks Python 3.10/3.12/3.13 and Linux/macOS/Windows, including an installed wheel outside the checkout.
+- The Bash uninstaller preserves checkouts and unrecognized commands; install refuses overwriting an unrelated command.
+- Clarified that the AI Agents Directory backlink supports a pending verification.
+
+## 0.53
+
+Elite QA pass: every claim verified, every path live.
+
+- Router repaired: accountable always equals the matched specialist; personal health/finance signals escalate to the owner; 66 routing probes green
+- `quality_os.py audit` repaired: scans all 61 employees across 12 departments, 0 gaps
+- All 61 capability contracts scrubbed to public identity; all manifests valid
+- Dead internal paths removed; canonical QA loop documented in `docs/QA-LOOP.md`
+- New: `fixtures/control-plane/` honest fixtures, `.github/workflows/ci.yml`, `docs/QA-LOOP.md`
+
+## 0.52
+
+
+Public edition of the private V5.2.2 workforce.
+
+Included:
+
+- 61 canonical employees under `employees/<department>/<name>/`
+- The existing chief of staff under `chief-of-staff/`
+- Four control-plane companions under `control-plane/`, so routing still runs
+
+Not included:
+
+- Alfred
+- Marketplace catalogs
+- Patent material
+- Newer employees and protocols
+- Eval receipts, fixtures, and private Git history
+
+Identity in public skills is Solaris Dev Shop. A model vendor is not the employee.

package/copilot/plugin/CODE_OF_CONDUCT.md

@@ -0,0 +1,5 @@
+# Code of conduct
+
+Be direct. Be accurate. Do not insult people, post other people's secrets, or pretend a sample is the whole library.
+
+Reports go to the repository issues after publication. Do not include credentials in a report.

package/copilot/plugin/CONTRIBUTING.md

@@ -0,0 +1,46 @@
+# Contributing
+
+Fixes to docs, broken paths, and secret leaks are welcome. New employees go through the [new employee proposal](.github/ISSUE_TEMPLATE/new-employee.md) issue first.
+
+New here? Follow the [contributor quickstart](docs/contributor-quickstart.md).
+Choose a concrete unclaimed task rather than the standing invitation itself.
+Open invitations and reviewed proposals are kept open so contributors can join;
+they are not unresolved product defects merely because they are open.
+
+Do not:
+
+- Add passwords, API keys, tokens, or real account details
+- Add employees that are not in the active roster (propose them via an issue instead)
+- Add Alfred, marketplace, or patent material
+- Rename a model vendor into the product or the employee
+
+Router changes must keep the suite green:
+
+```bash
+python3 -m unittest discover -s tests -v
+python3 control-plane/meta_control_plane.py self-test
+```
+
+Open an issue with the command you ran, what you expected, and what you observed. A pull request should do the same in the description.
+
+After editing a skill identity, summary or tags, run
+`python3 tools/build_skill_catalog.py` and commit the updated catalog.
+CI checks source paths and roster IDs to prevent stale discovery results.
+
+Validate `SKILL.md` discovery metadata after changing a skill:
+
+```bash
+python3 -m pip install -r requirements-dev.txt
+python3 tools/validate_skill_metadata.py
+```
+
+This developer check parses YAML and checks names and descriptions for all 61
+specialists plus the coordinator. Quote descriptions containing colons. Keep
+descriptions within 1024 characters and use words for comparisons. Detailed
+procedures belong in the body. The intake runtime remains standard-library only.
+
+New specialists are welcome. The active roster is 61 today; it can grow after
+proposal review. Start with the missing user outcome, the closest existing skill,
+and a small synthetic worked example. Do not invent experience or test results
+to fill a template. The [skill checklist](docs/new-skill-template/README.md)
+describes the source files and routing changes required after approval.