ghostchimera 0.4.0b0__tar.gz

ghostchimera-0.4.0b0/CHANGELOG.md

@@ -0,0 +1,227 @@
+# Changelog
+
+## 0.4.0-beta - 2026-05-13
+
+### Added
+
+- Personal MiniMind service with persisted admin consent, system-spec capture, approved file/email bootstrap, local-memory dataset generation, and primary-model RAG handoff prompts.
+- Whole-machine and email-artifact crawl consent scopes with configurable crawl roots, custom exclusions, default system/cache/dependency exclusions, and file/email limits.
+- Ghost Console MiniMind tab for consent, source paths, crawl toggles, bootstrap, revocation, status, and handoff generation.
+- Console API routes under `/api/console/minimind/personal/*`.
+- SDK helpers: `enable_personal_minimind()`, `bootstrap_personal_minimind()`, `personal_minimind_status()`, `minimind_handoff()`, and `revoke_personal_minimind()`.
+- CLI actions: `ghostchimera minimind personal-status`, `personal-consent`, `personal-bootstrap`, `personal-handoff`, and `personal-revoke`.
+- Personal MiniMind privacy/operation documentation covering broad crawl behavior, email artifact discovery, local storage, and local quantized runtime options.
+- Competitive capability intelligence layer with `ghostchimera capabilities`, `/api/console/capabilities`, a dashboard Capabilities tab, `docs/COMPETITIVE_CAPABILITY_MATRIX.md`, and a `competitive` eval suite.
+- Automated PR/diff review with `ghostchimera review-pr`, `/api/console/review-pr`, a dashboard Review tab, deterministic security/release heuristics, Markdown/JSON output, and competitive eval coverage.
+
+### Changed
+
+- Bumped package metadata to `0.4.0-beta`.
+- MiniMind personal bootstrap can generate training data from the full local memory corpus while preserving explicit consent gates.
+- Release readiness now gates the competitive matrix against Codex, Claude Code, LangGraph, CrewAI, Hermes-style tool gateways, and OpenClaw-style local autonomy benchmarks.
+- Competitive capability scoring now requires all tracked beta surfaces to be complete before the suite passes.
+
+## 0.3.0-beta — 2026-05-12
+
+### Added
+
+- **Chimera Pilot v2 runtime components**: `AIAgent` multi-turn loop
+  (`agent_loop.py`), `ContextCompressor` (token-budget context management),
+  `MCPWrapper` (JSON-RPC MCP connector), `CredentialPool` + `ExternalAuthProvider`
+  (OpenClaw-style credential injection), `ErrorClassifier` + `AutoRecoveryPlan`
+  (typed recovery taxonomy), `CheckpointManager` (replay-safe run snapshots),
+  `ToolsetManager` (named toolset composition), `SubagentPool`, `MixtureOfAgents`
+  (MoA scoring + Jaccard strategy selection), `BatchRunner`, `CronScheduler`, and
+  `GatewayServer` (WebSocket gateway with HTTP route registry).
+- **Gemini / Google AI Studio integration**: `GeminiProvider` (chat, long-context,
+  multi-agent history), `GeminiBackend` (REASONING + LONG_CONTEXT_DOC tasks), and
+  `ModelCatalog` entries with 1 M-token context models.
+- **Simulation backend** (`SimulationBackend`): kinematics trajectory planner,
+  digital-twin sensor emulation, and policy-test episode runner with collision
+  detection.
+- **Analytics and data pipeline backend** (`AnalyticsBackend`): count/sum/avg
+  group queries, simple linear-trend forecasting, z-score anomaly detection, CSV
+  parsing, schema validation, and knowledge-graph triple extraction.
+- **Document ingestion** (`DocumentIngester`): text chunking, CSV row ingestion,
+  duplicate-safe insert via `add_document_once`.
+- **DPI / LobsterTrap safety layer**: `BuiltinDPIEngine` (prompt injection,
+  credential-leak, PII, exfiltration, and intent-mismatch scanning),
+  `LobsterTrapProvider` wrapper, `LobsterTrapConfig.from_env()`, and
+  `SecurityMonitor` with per-category event aggregation and threat summary.
+- **OpenClaw modularity gaps closed** (all 10): `MediaProviders` (Gap 1),
+  `ApprovalHandler` + new `HookName` entries (Gap 2), `ToolMiddlewareChain`
+  (Gap 3), `Skill.check_requirements()` (Gap 4), `BackgroundService` +
+  `ServiceRegistry` (Gap 5), `HttpRouteRegistry` in gateway (Gap 6),
+  `PluginManifest` (Gap 7), `ExternalAuthProvider` in `CredentialPool`
+  (Gap 8), `SSRFPolicy` + `PilotPolicy.allowed_hosts` (Gap 9),
+  `TEXT_PROVIDERS` + `register_text_provider` (Gap 10).
+- **Red-team eval suite** (`redteam`): 9 cases covering prompt injection blocking,
+  credential-leak blocking, PII detection, exfiltration blocking, intent-mismatch
+  flagging, benign-prompt pass-through, `LobsterTrapProvider` enforcement,
+  `SecurityMonitor` aggregation, and `LobsterTrapConfig.from_env()`.
+- **Track 2 eval suite** (`track2`): 8 Gemini integration cases.
+- **Track 3 eval suite** (`track3`): 6 simulation/robotics cases.
+- **Track 4 eval suite** (`track4`): 9 analytics and data-pipeline cases.
+- **Coverage eval suite** (`coverage`): SSRF policy, approval token, material
+  policy, error classifier, MoA scoring, context compressor, autonomy queue,
+  checkpoint save/restore, and telemetry export format.
+- **Workspace eval suite** (`workspace`): 6 cases covering workspace context
+  injection into compiled tasks, no-injection on irrelevant objectives, freshness
+  score and citation quality per search result, empty-index graceful degradation,
+  `count()` tracking, and local-model profiles CLI.
+- **Memory store depth**: `MemoryStore.search()` returns `freshness_score`
+  (exponential decay from `created_at`, 30-day half-life), `citation_quality`
+  (freshness × content-length heuristic), and `created_at` per result. New
+  `stale_after_days` query-time filter and `count()` method.
+- **Workspace feedback loop**: `OperatorWorkspaceStore.workspace_context_for_objective()`
+  lightweight in-memory relevance retrieval. `ChimeraPilotKernel` accepts a
+  `workspace_store` parameter and injects matching context into
+  `TaskSpec.constraints["workspace_context"]` at compile time.
+- **Local model bootstrap CLI**: `ghostchimera local-model` subcommand with
+  `check`, `guide`, and `profiles` actions — reports system resources vs profile
+  requirements, step-by-step download instructions, and fit analysis.
+- **Production isolation guidance**: `docs/PRODUCTION_ISOLATION.md` — container
+  and VM hardening, state backup/restore, audit log retention, secret handling,
+  SSRF + network controls, and rollback runbooks.
+- **Material policy** (`MaterialPolicyEngine` + `MaterialRegistry`): temporal and
+  factual claim classification, adversarial prompt detection, and security-check
+  trace IDs.
+- **Runtime specialization warmup**: `detect_runtime_environment` and
+  `warm_runtime_specialization_cache` for pre-computing llama.cpp launch manifests.
+- **Async executor** (`executor_async.py`) and **parallel executor**
+  (`executor_parallel.py`) with cooperative cancellation and structured
+  failed-result return.
+- **Calibration async** (`calibration_async.py`) with concurrent backend probing.
+- New eval suites produce per-suite KPIs and release-gate summaries (all suites:
+  smoke, safety, autonomy, user-journey, coverage, redteam, track2, track3,
+  track4, workspace).
+
+### Changed
+
+- Scheduler strategy selection extended with `moa` strategy and `select_strategy`
+  supporting historical success rate and uncertainty inputs.
+- `PilotPolicy.autonomy_profile.cap_strategy` enforces per-profile ceilings on
+  strategy selection.
+- `MemoryStore.search()` signature now includes `stale_after_days` and
+  `freshness_half_life_days` optional keyword arguments (backward-compatible).
+- Blueprint scorecard updated: Real memory/retrieval 95%, Local model runtime 90%,
+  Conscious workspace 95%, Evaluation harness 90%, Production packaging 90%.
+
+---
+
+## 0.2.1 — 2026-05-09
+
+### Added
+
+- **Workspace sync quality flags**: `ghostchimera workspace sync-memory` now
+  reports low-confidence filtered records and marks stale (>30 days) and
+  conflicting records with quality metadata before they enter retrieval.
+- **Console operator routes**: browser console exposes evidence-add,
+  reflection-add, goal-update, and workspace-sync-memory API routes in addition
+  to workspace show.
+- **Readiness runbook route**: `/api/console/readiness` exposes the full release
+  gate checklist as a local operator runbook view.
+- **Autonomy job center**: `/api/console/autonomy/jobs` records profile-gated
+  preview runs; `/api/console/autonomy/schedules` manages disabled recurring
+  autonomy jobs via `CronScheduler`.
+- **User-journey eval suite** (`user-journey`): 6 cases covering top-level CLI
+  dispatch, config/state reporting, console operator routes, workspace sync,
+  workspace quality flags, and readiness runbook output.
+- **Workspace sync CLI** (`ghostchimera workspace sync-memory`): promotes
+  high-confidence evidence/reflections into the SQLite CWR store with provenance
+  metadata and duplicate-safe inserts.
+- **`ghostchimera doctor --production`**: production readiness report via
+  `production_readiness_report()`.
+
+### Fixed
+
+- `OperatorWorkspaceStore.save()` now uses atomic rename to prevent partial writes
+  on Windows.
+- Desktop backend kill-switch check respects the configured stop-file path.
+
+---
+
+## 0.2.0 — 2026-05-06
+
+### Added
+
+- **Local operator console** (`ghostchimera console`): WebSocket-backed localhost
+  UI for status, autonomy profile control, safe objective runs, optional browser
+  workspace controls, and release-readiness checks. Supports `--host`, `--port`,
+  `--http-port`, `--state-dir`, and `--no-open`.
+- **Operator workspace UX**: `ghostchimera workspace show` exposes inspectable
+  local self-model, working memory, attention ranking, goals, goal state, and
+  uncertainty summary. `operator_workspace.json` persists under the state
+  directory.
+- **Conscious workspace feedback loop**: `ghostchimera workspace sync-memory`
+  and `/api/console/workspace/sync-memory` promote high-confidence workspace
+  evidence and reflections into the SQLite CWR store with provenance metadata
+  and duplicate-safe inserts.
+- **Desktop control backend** (`DesktopRuntimeBackend`): dry-run default,
+  explicit live-mode opt-in, action-class filtering, kill-switch file, and
+  screenshot/action-log telemetry. `ghostchimera desktop-stop` for immediate
+  session halt.
+- **Autonomy profiles**: `assist`, `supervised` (default), `autonomous`, and
+  `generalist`. Configurable via `GHOSTCHIMERA_AUTONOMY_LEVEL` or
+  `ghostchimera autonomy set`. Each profile caps strategy, budget, parallelism,
+  and approval gates.
+- **True autonomy desktop toggle** in Ghost Console: when enabled, default run
+  uses possess-mode live desktop backend with expanded permissions.
+- **Runtime state and checkpointing**: `run_id`, `attempt_id`, and `checkpoint_id`
+  propagate through executor transitions; terminal-state checkpoint recording
+  connected to telemetry and replay bundle generation.
+- **Interrupt and cancellation**: cooperative cancellation for executor and
+  parallel execution entry points; cancelled parallel runs return structured
+  failed executions.
+- **Adaptive scheduler learning loop**: score breakdowns, configurable weights,
+  and bounded adaptation; outcome persistence wired through memory store;
+  strategy selection supports `single`, `fallback_chain`, `parallel`, and `moa`.
+- **Delegation and shared-state arbitration**: delegation contract primitives,
+  file lease arbitration, and structured merge conflict reports in
+  `ghostchimera.tool_layer.file_system`.
+- **External skill discovery**: `SkillRegistry` discovers bundled + workspace
+  skills from `~/.ghostchimera/skills/<name>/skill.py` (override via
+  `GHOSTCHIMERA_SKILLS_DIR`). `AgentCore` accepts `skill_registry` param.
+- **Production deployment mode** (`GHOSTCHIMERA_DEPLOYMENT_MODE=production`):
+  blocks shell, file writes, and live desktop unless external isolation,
+  security review, and human approval are all declared. `ProductionGuardrails`
+  and `production_readiness_report()`.
+- **Replayable observability**: replay bundles include run, decision, attempts,
+  transitions, and trace hashes; telemetry exports JSON/CSV and replay-bundle
+  files.
+- **Ghost mode taxonomy**: `whisper`, `shadow`, `possess`, and `ghost` modes with
+  per-mode policy implications.
+- **CI workflow**: `.github/workflows/ci.yml` with lint, tests, build, and wheel
+  smoke steps on push and pull request.
+- **Dockerfile** for reproducible container deployments.
+
+### Changed
+
+- `AgentCore.handle_request()` now tries Chimera Pilot first, falling back to
+  legacy planner/skill executor only when no backend can handle the task.
+- `PilotPolicy` split from `ExecutionPolicy`: scheduler decides *where* to run,
+  policy decides *whether* it is allowed — two separate boundary layers.
+- `ghostchimera doctor` extended with `--production` flag for production
+  readiness reporting.
+
+---
+
+## 0.1.0 — 2026-04-29
+
+Initial alpha release package.
+
+### Added
+
+- Chimera Pilot task IR, backend registry, scheduler, calibration, executor, verifier, telemetry, and CLI.
+- Conservative policy defaults for network and local Python/test execution.
+- Hardened Python runtime backend with bounded timeout, minimal environment, isolated interpreter mode, temporary cwd, and static rejection of high-risk calls.
+- Optional pyqpanda3 quantum simulator backend.
+- Policy-gated AgentCore shell, filesystem, and browser execution with audit records.
+- SQLite FTS local CWR memory retrieval backend and memory CLI commands.
+- Local model profiles, minimind-compatible provider contract, and optional llama.cpp/GGUF backend.
+- Ghost-native MiniMind architecture metadata, runtime inspection, attribution, and optional Transformers/PyTorch adapter.
+- Conscious workspace primitives for inspectable self-model, working memory, attention, and reflection state.
+- Smoke and safety eval harness with `ghostchimera-eval`.
+- Typed environment configuration and `ghostchimera --config-show`.
+- Source package metadata, MIT license, release docs, CI workflow, and release validation script.
+- Targeted unittest coverage for scheduling, fallback, policy gating, calibration, compilation, Python execution, local retrieval, local model profiles, evals, configuration, and release metadata.

ghostchimera-0.4.0b0/CHIMERA_PILOT.md

@@ -0,0 +1,103 @@
+# Chimera Pilot
+
+Chimera Pilot is Ghost Chimera's resource orchestration layer. It turns high-level objectives into neutral `TaskSpec` objects, chooses a backend, enforces policy, executes with fallback, verifies output, and records telemetry.
+
+## Why it exists
+
+The subsystem re-iterates quantum/classical resource-orchestration ideas for practical everyday agent infrastructure:
+
+- local Python runtimes;
+- model providers;
+- retrieval systems;
+- browser/search tools;
+- MCP-style tool servers;
+- optional quantum simulators.
+
+It does not require a quantum computer.
+
+## Governed Enterprise Change Agent Workflow
+
+When used as the product control plane, Chimera Pilot drives one enterprise change loop:
+
+1. Ingest repo/document objective context.
+2. Compile objective to normalized `TaskSpec` units.
+3. Schedule and execute with fallback/parallel strategy ceilings based on autonomy profile.
+4. Enforce policy and production guardrails.
+5. Verify and emit confidence-bearing, audit-ready outputs for change review.
+
+This shared flow is reused across Agentic Olympics, enterprise/security, IBM Bob, and AI GENESIS hackathon demos with track-specific framing.
+
+## Commands
+
+```bash
+chimera-pilot status --include-deterministic-backend
+chimera-pilot compile "retrieve memory about project"
+chimera-pilot calibrate --include-deterministic-backend
+chimera-pilot run "retrieve memory about project" --include-deterministic-backend
+chimera-pilot model-profiles
+```
+
+Use the local Conscious Workspace Retrieval store:
+
+```bash
+chimera-pilot memory-add --memory-db .ghostchimera-memory.sqlite3 --source project-note --content "Ghost Chimera stores local project memory."
+chimera-pilot memory-search --memory-db .ghostchimera-memory.sqlite3 "project memory"
+chimera-pilot run "retrieve project memory" --memory-db .ghostchimera-memory.sqlite3 --include-deterministic-backend
+```
+
+Python execution is denied by default:
+
+```bash
+chimera-pilot run "python: print(2 + 3)"
+```
+
+For trusted local code only:
+
+```bash
+chimera-pilot run "python: print(2 + 3)" --allow-python
+```
+
+Desktop control is denied by default:
+
+```bash
+chimera-pilot run "click submit button" --enable-desktop-backend
+```
+
+For explicit desktop-control opt-in and possess mode:
+
+```bash
+chimera-pilot run "click submit button" --enable-desktop-backend --allow-desktop-control --ghost-mode possess
+```
+
+Compiler shorthand prefixes are supported for desktop intent strength:
+
+```bash
+chimera-pilot run "dryrun desktop: click submit button" --enable-desktop-backend --allow-desktop-control --ghost-mode possess
+chimera-pilot run "live desktop: click submit button" --enable-desktop-backend --enable-live-desktop --allow-desktop-control --ghost-mode possess
+```
+
+## Built-in task kinds
+
+- `reasoning`
+- `code_edit`
+- `test_run`
+- `web_research`
+- `file_analysis`
+- `rag_query`
+- `tool_call`
+- `python`
+- `quantum_sim`
+- `desktop_control`
+
+## Built-in backends
+
+- `DeterministicBackend` for CI, smoke checks, and fallback tests.
+- `PythonRuntimeBackend` for trusted local Python/test execution.
+- `PyQPanda3Backend` for optional pyqpanda3 quantum simulation when installed.
+- `CWRBackend` for SQLite-backed local memory retrieval.
+- `LlamaCppBackend` for optional GGUF reasoning when a local model path is provided.
+- `DesktopRuntimeBackend` for dry-run desktop control (with explicit live constraints for real UI mutation).
+
+## Security
+
+Local execution is policy-gated. See `SECURITY.md` before enabling Python, shell, filesystem mutation, or network tasks.

ghostchimera-0.4.0b0/CONTRIBUTING.md

@@ -0,0 +1,39 @@
+# Contributing
+
+Thank you for contributing to Ghost Chimera.
+
+## Development setup
+
+```bash
+python -m venv .venv
+source .venv/bin/activate
+python -m pip install -e '.[dev]'
+python scripts/validate_release.py
+```
+
+## Engineering rules
+
+- Keep changes small and directly tied to the issue.
+- Do not refactor unrelated modules while fixing a bug.
+- Add tests for new behavior.
+- Keep local execution capabilities opt-in and policy-gated.
+- Do not add proprietary code, private APIs, scraped licensed artifacts, or copied implementation details from external products.
+
+## Test requirements
+
+Before opening a pull request, run:
+
+```bash
+python -m unittest tests.test_chimera_pilot tests.test_release_package -v
+python -m compileall ghostchimera tests
+python scripts/validate_release.py
+```
+
+## Clean-room requirements
+
+When implementing ideas inspired by external systems:
+
+1. Use public docs, papers, and open-source repositories only.
+2. Convert product-specific ideas into neutral architecture notes.
+3. Implement from the neutral spec, not from proprietary code.
+4. Add attribution or clean-room notes where appropriate.

ghostchimera-0.4.0b0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Fernando Garza
+
+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.

ghostchimera-0.4.0b0/MANIFEST.in

@@ -0,0 +1,5 @@
+include README.md LICENSE NOTICE SECURITY.md CONTRIBUTING.md CHANGELOG.md CHIMERA_PILOT.md
+recursive-include docs *.md
+recursive-include scripts *.py
+recursive-include tests *.py
+include ghostchimera/py.typed

ghostchimera-0.4.0b0/NOTICE

@@ -0,0 +1,29 @@
+Ghost Chimera
+
+Ghost Chimera includes Ghost-native MiniMind architecture metadata and optional
+runtime adapter behavior derived from public MiniMind project documentation and
+interfaces.
+
+MiniMind source project: https://github.com/jingyaogong/minimind
+Source commit reviewed: dddedc688121028dd8adab55b95d139ecd87205c
+MiniMind license: Apache License 2.0
+
+Ghost Chimera does not bundle MiniMind model weights. Local MiniMind inference
+requires operator-provided weights and optional runtime dependencies.
+
+Native Absorption
+
+Ghost Chimera includes Ghost-native implementations inspired by the author's
+adjacent projects:
+
+- chimeralang-mcp: confidence gating, tamper-evident handoff, and
+  query-aware context compression ideas.
+- ChimeraLang: belief confidence, variance guard, and operational reasoning
+  trace concepts.
+- OpenDrop: local model source resolution, hardware posture, model inventory,
+  license posture, and quantization recommendation concepts.
+- OpenChimera_v1: MCP capability normalization and operator sandbox journey
+  patterns.
+
+These capabilities are implemented inside Ghost Chimera and do not require
+those repositories as runtime dependencies.