Agent Lifecycle Kit v1.53.0

• Deterministic risk-aware execution profiles derive provider-neutral routes and S0/S1/S2 token, invocation, and wall-time caps.
• start --risk --risk-profile-out projects a read-only profile; workflow task-start --risk-profile explicitly authorizes it with strict lineage validation.
• S1/S2 task results require host-attested usage and fail closed on missing, estimated, or over-cap evidence.
• English and Russian operator documentation, package pins, plugin manifests, and marketplace references are updated to 1.53.0.

Agent Lifecycle Kit v1.52.0

Unified lifecycle start

• Adds agent-lifecycle start as the common entrypoint for task text/files, explicit frozen implementation, and stored ALK session resume.
• Keeps auto, research, plan, and review non-executing and review-gated.
• Requires complete frozen bindings for explicit --mode implement.
• Adds the stable agent-lifecycle-start-receipt.v1 contract and synchronized English/Russian documentation.
• Updates all package, plugin, and marketplace publication surfaces to 1.52.0.

Validated on Python 3.11-3.14 across Linux, macOS, and Windows. Independent implementation audit completed through OpenCode with no Claude/Anthropic usage.

Agent Lifecycle Kit v1.51.0

Python 3.14 and PyPI qualification

• Add supported Python 3.14 metadata and CI coverage across Linux, macOS, and Windows while retaining Python 3.11-3.13.
• Build and install wheel and source distributions in separate clean Python 3.14 environments during release qualification.
• Emit portable combined packaging evidence for both artifact types without raw process output.
• Document the official PyPI package and exact agent-lifecycle-kit==1.51.0 installation path in English and Russian.
• Update package, plugin, marketplace, documentation, and publication metadata to 1.51.0.

Validation: 779 tests on Python 3.11 and 3.14 locally, full Python 3.11-3.14 GitHub matrices on Linux/macOS/Windows, neutrality checks, offline release candidate, packaging smoke, and independent OpenCode Review Mesh audit.

Agent Lifecycle Kit v1.50.2

Security and architecture hardening

• Block generic descriptor-driven host launches and wildcard environment forwarding until a separately qualified local-launch path exists.
• Add shared receipt redaction, fail-closed neutrality completeness checks, descriptor installation metadata, and portable release-evidence validation.
• Split the CLI dispatcher into bounded routing modules without changing the public command contract.
• Normalize packaging evidence across POSIX and Windows paths.
• Update package, plugin, marketplace, documentation, and publication metadata to 1.50.2.

Validation: 776 tests, Linux/macOS/Windows CI on Python 3.11-3.13, neutrality checks, offline release candidate, packaging smoke, and independent OpenCode Review Mesh audit.

v1.50.1

Enables the PyPI publish pipeline (Trusted Publisher workflow); version-only bump, no functional changes.

v1.50.0

Release 1.50.0 adds deterministic Bug Forensics advisory refinement.

Highlights:

• Adds an advisory-only Bug Forensics policy for defect, regression, flaky failure, incident and security-bug task signals.
• Embeds bugForensicsAdvisory in adapter task intake receipts without activating workflow gates.
• Adds EN/RU Bug Forensics workflow guides and updates cookbook, template and skill guidance.
• Updates package and plugin publication metadata to 1.50.0 / v1.50.0.

Validation:

• Full unittest discovery: 736 tests passed locally.
• Focused advisory, workflow gate, CLI and docs tests passed.
• No-model-call, docs compatibility, publication version and packaging smoke gates passed locally.
• GitHub Actions: 12/12 checks passed on PR #77.

v1.49.0

Release 1.49.0 adds a compact read-only goal/progress view for active lifecycle runs.

Highlights:

• Added agent-goal-progress-view.v1 and agent-lifecycle goal view.
• Added terminal rendering for goal/progress summaries while keeping JSON output as the default contract.
• Updated EN/RU quickstarts, lifecycle cookbook, CLI and read-only status references.
• Updated package and plugin publication metadata to 1.49.0.

Validation:

• Focused goal/progress/docs tests passed.
• No-model-call, docs compatibility and publication version gates passed.
• Full unittest discovery passed locally: 735 tests.
• Packaging smoke passed in a local build venv.
• GitHub Actions PR #76 passed all 12 jobs.

v1.48.0

Release 1.48.0 adds an optional external memory bridge.

Highlights:

• Added agent-external-context-import-receipt.v1 and validation.
• Added context external-import for explicit local memory/context files.
• Added context episode-retrieve with optional non-proof externalContextHints.
• Documented EN/RU safety boundaries for external memory and production resource use.
• Updated plugin and marketplace metadata to 1.48.0 / v1.48.0.

Validation:

• Focused contract/context/CLI/docs tests passed.
• No-model-call scan, docs compatibility, publication version validation, full unittest discovery, packaging smoke, and GitHub CI passed.

v1.47.0

Release 1.47.0 adds adapter-owned event capture guidance.

Highlights:

• Added EN/RU event capture matrix for all adapters.
• Added Codex, Claude Code, and OpenCode event bridge examples.
• Added validator for declared event capture routes.
• Updated plugin and marketplace metadata to 1.47.0 / v1.47.0.

Validation:

• Focused event guidance and docs tests passed.
• Docs compatibility, publication version validation, full unittest discovery, packaging smoke, and GitHub CI passed.

v1.46.0

Release 1.46.0 adds Review Mesh operator templates and the review-mesh prepare helper.

Highlights:

• provider-neutral templates for leader draft review, parallel research synthesis, and implementation-audit panels
• review-mesh template-list and review-mesh prepare for local profile, assignment packet, and prepare receipt generation
• replaceable Codex, Claude Code, and OpenCode/GLM examples documented without storing concrete provider/model claims in portable contracts
• package and plugin publication metadata updated to 1.46.0 / v1.46.0

Validation:

• focused Review Mesh/docs tests passed
• no-model-call scan for operator_templates.py passed
• docs compatibility and publication version checks passed
• full unittest discovery passed: 722 tests
• packaging smoke passed