Repository navigation
Releases: AidALL/ghost-alice-autopilot
Release list
Ghost-ALICE Autopilot 0.5.1
Ghost-ALICE Autopilot 0.5.1 records Core-authenticated completion evidence and selects eligible tasks from an approved continuation plan. Its adapters use one selected Core runtime; they do not repeat Core's semantic or authority judgment.
Requires Ghost-ALICE Core 0.5.1+ and Python 3.11+. Install from the Core checkout with bash install.sh --platform codex --addon autopilot or select --platform claude.
Validation covers the canonical project tests, Core integration and isolated installed providers. Publication, criterion application, task dispatch and actual host execution remain distinct observations.
Ghost-ALICE Autopilot 0.5.0
Ghost-ALICE Autopilot 0.5.0
Requires Ghost-ALICE Core 0.5.0. Registered pretool/Stop adapters prepare capture, publish Core-reviewed proof, apply qualified criteria and dispatch separately admitted plan tasks. Accepted proof preserves original sources, times and dependency mappings.
Install through the Core installer with bash install.sh --addon autopilot, or .\install.cmd --addon autopilot on Windows. See the current package documentation and compatibility scope.
Validation covers local Python 3.11 unit and isolated Codex/Claude adapter tests plus the linked Ubuntu CI results. Native Windows behavior and fresh model-behavior acceptance remain outside those observations. This release does not assert complete platform compatibility.
Ghost-ALICE Autopilot 0.4.1
Ghost-ALICE Autopilot 0.4.1 is paired with Core 0.4.1. The technical minimum remains Core 0.4.0.
From a cloned Core repository, install the coordinated pair with:
bash install.sh --addon autopilot- Select the firing session's run without adopting foreign-session or project-paused state.
- Preserve the selected project when derived storage is unavailable; retain explicit run-directory authority.
- Handle derived permission and read-only filesystem failures without hiding unexpected storage errors or explicit-target failures.
- Bind approval generation, input identity, completion receipts and original evidence to the selected run.
- Isolate subprocess fixtures and test exact-session selection, read-only paths, publication admission and the release inventory.
- Synchronize current bilingual installation, usage, approval, pause/stop and recovery guidance.
The complete local suite passed with 471 tests and 741 subtests. Repository CI verifies this release with the integrated Core main. Fresh native Windows and Claude inference evaluations, universal reliability and quantified token savings are outside the verified scope.
Ghost-ALICE Autopilot v0.4.0
Ghost-ALICE Autopilot v0.4.0 Release Notes
Language: English | Korean
Changes
- Move approved runs, tasks, events, consumed inbox receipts and approval history to SQLite. Bound runs share Core transactions; standalone runs retain a separate database.
- Bind admission and evidence to exact session/input identity, criterion definitions and approval generation. Preserve progress, budget and OFF for the same approved contract; archive superseded generations.
- Capture prospective provenance before tools, surface current preparation commands once, and publish unchanged proof before the initial final answer. Original source, time and proof digest remain bound through recovery.
- Keep explicit current-input reapproval attached to the user task instead of substituting an advisory conduct plan. Recovery reuses valid business evidence and preserves the user's requested answer.
- Ignore unrelated-session databases during pre-tool provenance discovery, while retaining errors for invalid current-session authority. Bind CI to the paired Core runtime and verify both registered hook events.
Compatibility and upgrade
Autopilot 0.4.0 requires Core 0.4.0 or newer for shared SQLite APIs. Update and reinstall both packages together. Older installers gaining adapter support in 0.2.2 does not make their runtime sufficient for this release. Stop old writers before migration and preserve a consistent backup; migrated JSON files are no longer authoritative. Internal schema versions remain independent of product versions.
Verification and limits
The delivered implementation's selected isolated evaluations observed initial completion publication, correct target and current-input handling, and no storage-only repeat inspection. Original failed attempts and evaluation limitations remain preserved. These observations do not establish universal reasoning quality or eliminate host UI duplication when a Stop hook rejects an answer.
Run the repository CI against the paired Core checkout. Existing platform evidence in compatibility-matrix.json is not relabeled as a fresh run for this version. Git tags and hosted release publication are separate from this source-version update.
License
The project remains open source under Apache-2.0.
Ghost-ALICE Autopilot v0.3.0
Ghost-ALICE Autopilot v0.3.0 Release Notes
Date: 2026-09-26
Language: English | 한국어
This release aligns the Autopilot product version with Ghost-ALICE core 0.3.0 and makes continuation depend on the selected session's current intent before pending state changes are applied. This file is the source for the GitHub release body; VERSION, the addon manifest, and the changelog identify the same product release.
What Changed
- Current intent is checked before consuming pending completion receipts, importing conduct plans, or changing a work queue. A changed objective or missing/conflicting selected context parks the run before those actions.
- Hosts using
agent-runtimemust explicitly provideGHOST_ALICE_PLATFORM=agent-runtime,GHOST_ALICE_SESSION_ID, and an absoluteGHOST_ALICE_SESSION_INTENT_ROOT. The selectedsession-intent-ledger.v1record must match that platform and session. Unknown platforms and invalid selected context do not borrow a native or sibling ledger. - Platform names supplied by the firing environment are normalized consistently before session-binding comparisons, so valid mixed-case names no longer park the run. Stored ledger and approval identities retain their strict validation.
- When the native platform is omitted, continuation selects the approved native platform and validates its actual current session. Malformed stored platform bindings stop continuation before receipts or queues change; approval does not supply a replacement current session ID.
- For a selected Codex context, explicit hook session identity takes precedence over the native thread, which takes precedence over a generic session value or shared pointer. This applies before creating a run and before continuing it. Claude ignores inherited Codex identity; unspecified host platforms retain the existing native discovery behavior.
- Ordinary refinements within an approved objective retain approval. Valid completion receipts are consumed once; reentry does not repeat the completed transition.
- English and Korean guidance now agrees on installation, continuation, compatibility, and the recommended release pair. Apache-2.0 licensing is unchanged.
Compatibility and Upgrade
Use Ghost-ALICE core 0.3.0 with Autopilot 0.3.0 for the coordinated intent-record and session-binding changes. The existing installer compatibility floor remains core 0.2.2; it is distinct from the recommended pair. Internal schema versions remain unchanged.
Run installation from the core checkout. This addon has no standalone root installer:
bash install.sh --addon autopilot --addon-tag v0.3.0
bash install.sh --statusThe official installer targets Claude Code and Codex. The explicit agent-runtime adapter contract lets another host supply session intent, but that host must still implement model access, tools, approval boundaries, and event dispatch. This release does not add those capabilities to arbitrary models by installation alone.
Verification and Limits
- Adapter state regression tests cover platform identity, selected intent context, changed goals, queue preservation, and single-use receipt handling. The firing environment accepts canonical, uppercase, or mixed-case platform names while stored ledger identity remains strict.
- Privileged adapter tests and compatibility tests cover the runtime boundary and release package contracts.
- Recorded-state replay used supplied and local intent records, but those private artifacts are not published with this release, so their numerical results are omitted here. Public test sources describe regression coverage, not general intent-understanding accuracy or a set of independent user sessions.
- This release's ledger validation did not include a fresh Claude model-inference run. Earlier live Claude/Codex evidence in the compatibility matrix is not relabeled as fresh
0.3.0evidence. Linux and Windows gaps in that matrix still prevent a full cross-platform compatibility claim.
See compatibility-matrix.json and the core release notes for the respective verification boundaries.
License
Ghost-ALICE Autopilot remains open source under Apache-2.0. The coordinated version does not change the license.
ghost-alice-autopilot 0.1.2
ghost-alice-autopilot v0.1.2 Release Notes
Date: 2026-08-17
Scope: v0.1.2 aligns the autopilot addon with Ghost-ALICE core v0.2.2, strengthens project-root and permission behavior, makes decision promotion status-aware, isolates all child scratch state, and prepares evaluator-visible diagnostics for separate hidden-purpose core evaluation.
Status: this file is the release-body source of truth for v0.1.2; the published release body must remain synchronized with this file and CHANGELOG.md.
Main Changes
- Raised the minimum Ghost-ALICE core version to
0.2.2and synchronized the addon and repository versions at0.1.2. - Rejected explicit relative project roots, ignored relative derived candidates, and kept permission failures fail-soft only for implicitly derived run directories and locks.
- Prevented
promote-decisionfrom writing actions when the effective decision is incompatible with the target work-item status, including ready-item reopen attempts and missing targets. - Added repository-local project runtimes and a canonical pytest launcher that owns temp, Python cache, clean child working directories, failure diagnostics, and successful cleanup.
- Preserved full continuation patterns without character-count truncation and resolved relative Codex last-message paths before clean-CWD execution.
- Kept Claude and Codex live diagnostic prompt transport symmetric while labeling this harness evaluator-visible; hidden-purpose installed behavior evidence is produced by the core blind controller.
- Replaced the fixed Claude live-diagnostic budget with a positive finite
--claude-max-budget-usdper-scenario cap whose default is1.00. - Unified the live scenario response contract and accepted one expected-key fenced JSON result around required governance control blocks while rejecting ambiguous candidates.
- Emitted live diagnostic JSON as UTF-8 under legacy Windows stdout encodings so Unicode model text cannot terminate an otherwise successful run.
- Removed display-width prose wrapping and shipped the same structural Markdown wrapping gate as core.
Verification Surface
- Run
python scripts/run_project_tests.py testsfor the complete addon suite and package contract. - Run the addon Markdown wrapping checker and compatibility-surface contract after all release files are in the Git index.
- Install core
v0.2.2plus addonv0.1.2for both Claude and Codex, then run evaluator-visible diagnostics and the separate core-owned blind cases.
Compatibility Boundary
- Ghost-ALICE core
v0.2.2is the minimum supported core for this addon release. - Claude Code and Codex are
verified-localthrough installed status, credentialed live semantic diagnostics, and five purpose-hidden fresh-session cases per platform. - Platform claims remain bounded by
compatibility-matrix.jsonand fresh installed-session evidence.
Release Boundary
- The Git tag, changelog section, release body, addon manifests, installed sidecars, and both platform adapters must identify
v0.1.2before publication is considered complete.
ghost-alice-autopilot v0.1.1
ghost-alice-autopilot v0.1.1 Release Notes
Date: 2026-07-01
Scope: published v0.1.1 changes after v0.1.0, including Codex live semantic smoke hardening, admitted acceptance-criterion bootstrap, validated criterion met-flips, reset-N autopilot resume budgeting, platform-neutral continuation signals, and invalid-decision quarantine.
Status: v0.1.1 is the public patch release for the Ghost-ALICE core v0.2.1 compatibility window. The local document is the release-body source of truth for the addon repository; if the GitHub Release body diverges, update the published body from this file.
Main Changes
- Hardened
scripts/live_semantic_e2e.pyso Codex live semantic runs detect config failures, command shim resolution issues, unsupported hook-trust flags, and missing prompt output without false pass/fail classification. - Fed live prompts through stdin for more reliable high-context Codex prompt execution.
- Added unittest coverage for command resolution, config failure classification, stdin prompt handling, and runtime flag compatibility.
- Added the live semantic unittest module to the tracked release package contract.
- Added a reset-N io-trace resume budget: an approved run may retry a decision-less running item for a bounded number of io-trace-backed stops, replenished only when the session-intent ledger advances.
- Added admitted acceptance-criterion bootstrap, replacing the previous io-trace-presence auto-attach heuristic (behavior change): the adapter no longer attaches to a session on io-trace material alone; it materializes a project-local
.autopilotrun only when session intent contains admitted, unmet acceptance criteria and explicit approval evidence exists. - Added validated criterion met-flips: a promoted
continue_nextdecision with criterion-bound[completion-check]evidence marks the corresponding admitted session-intent criteria as met through the core ledger API when that API is reachable. - Rendered Stop-hook continuation signals with portable project/home-relative paths for Bash rows, source locators, and allowed surfaces while preserving absolute/raw values in stored state and local audit logs.
- Quarantined invalid promoted consistency-decision files to
consistency-decision.rejected.jsonbefore preserving the fail-closed validation error, preventing permanent re-raise loops. - Hardened the Stop adapter against cross-run/session contamination by binding run-state continuation to the approved project/run surface and rejecting foreign promoted decision material.
Verification Surface
- Local unit verification should run
python -m unittest discover -s tests -p "test_*.py". - GitHub Actions uses the same
python -m unittest discoveraddon suite instead of a pytest wrapper. - Focused adapter verification should include
tests/test_autopilot_state.py,tests/test_autopilot_messages.py, andtests/test_autopilot_session_bridge.py. - Live semantic verification should run
python scripts\live_semantic_e2e.py --runtime codex --scenario-source static --execute --jsonlagainst the installed core plus addon runtime. A semantic mismatch, hook-incomplete run, timeout, auth/config blocker, or inference failure is a non-zero harness result. - GitHub Actions passed before publication of
v0.1.1; future patch edits must rerun the same unittest and live semantic surfaces before updating the published release body.
Compatibility Boundary
- Codex live semantic smoke is the verified runtime target for this patch release.
- Ghost-ALICE core
v0.2.1or newer is the supported core floor for thev0.1.1addon contract. Older core versions may lack the runtime-core audit, ledger met-flip, and privileged-adapter install semantics required by this release. - Claude support remains bounded by the compatibility matrix and does not become a full live compatibility claim through this release note.
- Linux and Windows support posture remains governed by
compatibility-matrix.json; this release hardens Windows command resolution but does not claim a full Windows live run unless that matrix says so.
Ghost-ALICE Autopilot v0.1.0
Ghost-ALICE Autopilot v0.1.0 — first release of the official autopilot-mode addon for Ghost-ALICE OS.
Autopilot is installed from the Ghost-ALICE core checkout (it has no standalone root installer):
bash install.sh --addon autopilot # macOS / Linux / WSL / Git Bash.\install.cmd --addon autopilot # Windows Command Prompt / PowerShellAdded
autopilot-modeprivileged adapter package with approval-gated.autopilot/state, task queue handling, continuation events, and Stop-event continuation payloads.- Session-intent bridge that can materialize approved
.autopilot/run state from Ghost-ALICE session intent ledgers after explicit approval. - Governance signal and promotion flow for diagnostic candidates, promoted consistency decisions, approved conduct plans, and candidate-only boundary checks.
- Compatibility matrix SSOT covering macOS, Linux, Windows shells, Claude Code, and Codex support posture.
- Fresh install E2E and local compatibility tests covering installed skill layouts, adapter smoke behavior, release package tracking, and documented support surfaces.
Changed
- Public docs state that install commands run from the Ghost-ALICE core checkout; this addon repository does not provide a standalone root installer.
- Compatibility matrix evidence describes the current support contract rather than retaining dated run logs.
- Release package tests check tracked release surfaces instead of shallow source-string fixtures.
Fixed
- Stop continuation reuses existing session-intent, io-trace, and governance signal receptors instead of stopping when unresolved runtime material exists.
- Candidate files remain diagnostic-only; adapter-consumable action files require explicit promotion.
- Claude support remains scoped as simulated-local until a credentialed Claude live semantic run passes.
Compatibility boundary
This release does not claim full cross-platform compatibility. Per compatibility-matrix.json:
- macOS, Codex: verified-local
- Claude Code: simulated-local (until a credentialed Claude live semantic run passes)
- Linux, Windows Command Prompt, Windows PowerShell 5/7: not-run
Requires Ghost-ALICE OS core v0.2.0 or later.