Harbor Self-Evolving 0.4.0

Harbor Self-Evolving 0.4.0 adds a first-class DeepSeek Harness Web experience while preserving the existing safe evaluation boundary.

Highlights

• Native Harbor conversation tab with Job status, Candidate identity, evaluation-context digests, metrics, exceptions, and Promotion decisions.
• Dedicated result cards for all four Harbor Agent tools.
• Harbor Evolution Settings/Doctor section for installation and runtime diagnostics.
• Original embedded ocean-and-whale visual with responsive and bilingual layouts.
• Same-origin, read-only, non-cacheable Host dashboard endpoint.
• Portable npm client bundle built for both registry publication and linked source development.

Evaluation, comparison, promotion, and deployment remain explicit Agent + Skill and CI/CD actions; refreshing the GUI cannot launch a Job or change the active Champion.

Packages

• npm: dsh-harbor-evolution@0.4.0
• PyPI: harbor-dsh-evolution==0.4.0

Harbor Self-Evolving v0.3.0

Plugin + Skill first

Harbor Self-Evolving is now positioned primarily as an installable DeepSeek Harness Plugin + Skill product. The repository examples remain reference implementations and are no longer the required starting point.

One-command setup

From a business Agent workspace:

npx dsh-harbor-evolution@0.3.0 setup --project-root "$PWD"

The setup command installs the matching Harbor Python adapter and DSH bundle, updates only the profile's harbor-evolution block, verifies the Harbor entry point, and prints the exact DSH restart command and first Skill invocation.

Improvements

• bundle the model- and user-invocable evolve-agent-with-harbor Skill with four deterministic evaluation tools
• preserve unrelated DSH profile patch entries and make repeated setup idempotent
• default to the web profile while supporting custom profiles, DSH homes, runtime directories, and Job directories
• unify public-registry and source-checkout installation
• document Candidate/Job comparability and the CI/CD promotion boundary

Packages

• npm: dsh-harbor-evolution@0.3.0
• PyPI: harbor-dsh-evolution==0.3.0

Verification

• 17 Python tests and package build passed
• 12 Node.js tests and npm package validation passed
• public one-command installation succeeded in a clean DSH profile and Python runtime
• Harbor discovered dsh-evolution, the bundled Skill was present, and DSH composed the expected profile configuration

Harbor Self-Evolving v0.2.0

What is new

• Bundles the project-maintained evolve-agent-with-harbor Skill through DSH's official Skill Registry.
• Gives Agents a stable workflow for requirements clarification, workspace initialization, baseline evaluation, evidence-driven optimization, regression comparison, and Promotion Gate reporting.
• Adds Candidate, Dataset, Verifier, and Promotion Policy initialization guidance.
• Adds a complete local DSH Web installation and first-run guide.
• Preserves the four deterministic Harbor tools and keeps deployment outside the evaluation boundary.

Install

uv pip install harbor-dsh-evolution==0.2.0
pnpm dlx @deepseek-ai/dsh@0.1.0-rc.6 \
  plugin --profile web add -w dsh-harbor-evolution@0.2.0

Restart DSH, then invoke:

/evolve-agent-with-harbor

Validation

• 17 Python tests
• 8 Node.js tests
• npm package installation in a clean DSH Web profile
• official DSH Skill Registry load and lifecycle verification

Full changelog: v0.1.0...v0.2.0

Harbor Self-Evolving v0.1.0

First public release of the Harbor-based Agent self-evolution template.

Highlights

• Snapshot immutable DeepSeek Harness Candidates with reproducible manifests.
• Evaluate Candidates through Harbor and preserve metrics, failures, ACP trajectories, and context digests.
• Compare Candidates with deterministic promotion gates before CI/CD promotion.
• Expose snapshot, run, result, and compare tools through a DSH Cordis bundle.
• Ship DeepResearch and shell examples for end-to-end validation.

Packages

• npm: dsh-harbor-evolution@0.1.0
• PyPI: harbor-dsh-evolution==0.1.0

Licensed under the MIT License.