https://istarwyh.github.io/harbor-self-evolving/docs/reference/Recent content in Reference on Harbor Self-EvolvingHugoen-USSun, 20 Sep 2026 10:28:22 +0800https://istarwyh.github.io/harbor-self-evolving/docs/reference/tools/Mon, 01 Jan 0001 00:00:00 +0000https://istarwyh.github.io/harbor-self-evolving/docs/reference/tools/<p>The Plugin exposes <strong>19 strict tools</strong>: ten workspace/Job mutations enter DSH one-shot approval; nine operations are read-only or in-memory. If the Host lacks the approval seam, mutation tools fail closed.</p> <div class="td-table-scroll td-table-scroll--static"> <table> <thead> <tr> <th scope="col">Tool</th> <th scope="col">Purpose and minimum boundary</th> <th scope="col">Mode / result</th> </tr> </thead> <tbody> <tr> <td><code>harbor_candidate_snapshot</code></td> <td>Freeze one Cordis composition as an immutable Candidate Manifest.</td> <td><strong>Writes local artifact · approval.</strong> Next: Dataset/Stack doctor or Context preview.</td> </tr> <tr> <td><code>harbor_model_binding</code></td> <td>Read the current DSH default provider/model/reasoning identity.</td> <td><strong>Read-only/in-memory.</strong> Returns a non-secret binding draft; never credentials.</td> </tr> <tr> <td><code>harbor_evolution_init</code></td> <td>Compile an accepted Dataset/Generator/Evaluator/Optimizer card into a non-overwriting Stack project.</td> <td><strong>Writes local artifacts · approval.</strong> Does not run evaluation.</td> </tr> <tr> <td><code>harbor_evolution_doctor</code></td> <td>Validate Candidate, Dataset, Stack, optional Policy and execution architecture before cost.</td> <td><strong>Read-only.</strong> Returns blocking diagnostics.</td> </tr> <tr> <td><code>harbor_quick_diagnostic_init</code></td> <td>Create one Query, minimal Host-model Candidate, runnable Task and non-promotion Evaluator.</td> <td><strong>Writes local artifacts · approval.</strong> Wiring diagnostic only.</td> </tr> <tr> <td><code>harbor_session_diagnostic_preview</code></td> <td>Preview 1–10 recent completed top-level Sessions in exact current workspace.</td> <td><strong>Read-only/in-memory.</strong> Safe metadata + 15-minute owner-bound selection token.</td> </tr> <tr> <td><code>harbor_session_diagnostic_run</code></td> <td>Revalidate a selection token, freeze a private redacted Batch and run one Trial per Session.</td> <td><strong>Starts evaluation · writes · approval.</strong> Historical, Gate N/A.</td> </tr> <tr> <td><code>harbor_dataset_validate</code></td> <td>Validate manifest, Task uniqueness, instructions, paths, sensitive metadata and source digest.</td> <td><strong>Read-only.</strong> Does not repair or rewrite identity.</td> </tr> <tr> <td><code>harbor_context_preview</code></td> <td>Refresh Candidate manifest, preview Context v3 and find comparable baselines.</td> <td><strong>Writes refreshed manifest · approval.</strong> No Job.</td> </tr> <tr> <td><code>harbor_eval_run</code></td> <td>Run strict diagnostic or promotion-eligible Candidate Job with frozen identities.</td> <td><strong>Starts evaluation · writes · approval.</strong> Returns Job identity.</td> </tr> <tr> <td><code>harbor_eval_result</code></td> <td>Read summary, Job, Dataset, progress, Trial or governance view.</td> <td><strong>Read-only.</strong> Bounded, recursively redacted, explicitly untrusted envelope.</td> </tr> <tr> <td><code>harbor_resolve_page_context</code></td> <td>Resolve exact-session opaque <code>@harbor</code> page context and current revision.</td> <td><strong>Read-only.</strong> Returns narrow metadata, typed refs and navigation action.</td> </tr> <tr> <td><code>harbor_get_evidence</code></td> <td>Read one Evidence item through an exact typed ancestry ref.</td> <td><strong>Read-only.</strong> Never accepts a guessed path/id.</td> </tr> <tr> <td><code>harbor_propose_action</code></td> <td>Draft one expiring Workbench action from an explicit user request and fresh context.</td> <td><strong>In-memory draft.</strong> Never writes resources, starts a Job, Gates or deploys.</td> </tr> <tr> <td><code>harbor_evaluator_inspect</code></td> <td>Inspect active descriptor, implementation kind, ternary Criteria and bounded editable source.</td> <td><strong>Read-only.</strong> Secret/local-path-shaped source is omitted.</td> </tr> <tr> <td><code>harbor_evaluator_update</code></td> <td>Replace one allowlisted source with optimistic concurrency and new Evaluator/Stack versions.</td> <td><strong>Writes local artifacts · approval.</strong> No automatic evaluation or Gate.</td> </tr> <tr> <td><code>harbor_ground_truth_init</code></td> <td>Create a non-overwriting independent Ground Truth draft with explicit provenance.</td> <td><strong>Writes local artifact · approval.</strong> Human/programmatic/consensus/model/external.</td> </tr> <tr> <td><code>harbor_evaluator_meta_evaluate</code></td> <td>Compare repeated observations with independent GT and emit ESF/SCE/RCR.</td> <td><strong>Writes report · approval.</strong> Evaluator governance, not Candidate promotion.</td> </tr> <tr> <td><code>harbor_candidate_compare</code></td> <td>Apply deterministic Promotion Gate to comparable Baseline and Candidate Jobs under Policy.</td> <td><strong>Writes Gate artifact · approval · promotion eligible.</strong> Never deploys.</td> </tr> </tbody> </table> </div> <h2 id="badges">Badge meanings </h2> <ul> <li><strong>Read-only:</strong> does not modify workspace evaluation state.</li> <li><strong>Writes local artifacts:</strong> creates or versions files in the bounded Harbor workspace.</li> <li><strong>Starts evaluation:</strong> can incur runner/Judge/model work after approval.</li> <li><strong>Promotion eligible:</strong> produces evidence or decisions usable by a promotion policy; Historical and quick diagnostic paths do not.</li> </ul> <p>Tool success means the declared operation completed—not that Candidate quality improved or production changed.</p>