opencode-harness-kit 0.5.6 → 0.5.7

package/CHANGELOG.md

@@ -12,6 +12,38 @@ is not happening.
 
 ---
 
+## [0.5.7] — 2026-10-21
+
+Published. The tag is what cut it: `git tag v0.5.7 && git push origin main
+v0.5.7`, and the publish job watches for exactly that.
+
+A patch, and a small one: the shipped profiles are refreshed against the
+live catalogue, and a test that pinned the catalogue's old answer stops
+pinning it.
+
+### Changed
+
+- **The profiles are refreshed against the live catalogue.**
+  `oc-harness refresh --all` is the kit's own command for this, and it had
+  not been run since the 19th. The catalogue now says that
+  `minimax-m3.1-flash-free` takes a temperature, and that
+  `longcat-2.5-preview-free` does not; the profiles said the opposite in one
+  case each. The `verified` date is the day the profile was last checked
+  against the catalogue, and leaving it stale after a check makes it a lie.
+
+  Nothing in the generated config changes: the kit sets no temperature at
+  all, so the profile's answer only decides what the header says.
+
+### Fixed
+
+- **A test failed on a day when nothing in the kit had changed.** The test
+  for "temperature is only emitted for models that support it" asserted
+  that one particular model does not support one - a value the catalogue
+  owns and changes. The test now asks whether the rule holds: no config the
+  kit writes may carry a temperature, and every profile has to say whether
+  its model takes one. A test that pins a value the world owns fails for
+  the wrong reason.
+
 ## [0.5.6] — 2026-10-21
 
 Published. The tag is what cut it: `git tag v0.5.6 && git push origin main
@@ -1437,7 +1469,8 @@ silent in the worst way:
 
 ---
 
-[Unreleased]: https://github.com/pham-canh/opencode-harness-kit/compare/v0.5.6...HEAD
+[Unreleased]: https://github.com/pham-canh/opencode-harness-kit/compare/v0.5.7...HEAD
+[0.5.7]: https://github.com/pham-canh/opencode-harness-kit/compare/v0.5.6...v0.5.7
 [0.5.6]: https://github.com/pham-canh/opencode-harness-kit/compare/v0.5.5...v0.5.6
 [0.5.5]: https://github.com/pham-canh/opencode-harness-kit/compare/v0.5.4...v0.5.5
 [0.5.4]: https://github.com/pham-canh/opencode-harness-kit/compare/v0.5.3...v0.5.4

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "opencode-harness-kit",
-  "version": "0.5.6",
+  "version": "0.5.7",
   "description": "Harness kit for OpenCode: capability profiles, a fleet of job lanes with one free model each, role-based reasoning effort, context pruning, and measured baselines.",
   "license": "MIT",
   "homepage": "https://github.com/pham-canh/opencode-harness-kit#readme",

package/profiles/fleet-all.json

@@ -5,7 +5,7 @@
   "provider": "opencode",
   "model": "minimax-m3.1-flash-free",
   "family": null,
-  "verified": "2026-10-19",
+  "verified": "2026-10-21",
   "retention": "may-train",
   "maxRetention": "may-train",
   "notes": "Every lane gets the model built for it: beam plans, Nemotron 3 Ultra reviews, Lightning sweeps, MiMo sees, GLM audits, Super VL reasons with the effort turned off. Five of the ten providers state that collected data may be used to improve the model, or log the session, so the profile as a whole is may-train and the generated config says so in its header. Install it for public or throwaway repositories.",
@@ -15,7 +15,7 @@
   },
   "supports": {
     "reasoning": true,
-    "temperature": false,
+    "temperature": true,
     "toolCall": true,
     "structuredOutput": false,
     "attachment": true

package/profiles/fleet-safe.json

@@ -5,7 +5,7 @@
   "provider": "opencode",
   "model": "minimax-m3.1-flash-free",
   "family": null,
-  "verified": "2026-10-19",
+  "verified": "2026-10-21",
   "retention": "zero-retention",
   "maxRetention": "zero-retention",
   "notes": "The default. Every lane resolves to a model whose provider states a zero-retention policy: no training, nothing kept. The cost is that six of the ten jobs run on a second-choice model - the planner loses beam (30-day retention) and the reviewer loses Nemotron 3 Ultra - so this is the floor, not the ceiling. The chains are ordered within the ceiling rather than copied from `fleet-all`: `locate` takes deepseek because it publishes a `low` effort level that longcat does not, and `vision` takes minimax because it reads an image in 7 seconds. `fleet-all` gives every lane the model built for it.",
@@ -15,7 +15,7 @@
   },
   "supports": {
     "reasoning": true,
-    "temperature": false,
+    "temperature": true,
     "toolCall": true,
     "structuredOutput": false,
     "attachment": true

package/profiles/longcat-2.5-preview-free.json

@@ -5,7 +5,7 @@
   "provider": "opencode",
   "model": "longcat-2.5-preview-free",
   "family": "longcat",
-  "verified": "2026-10-19",
+  "verified": "2026-10-21",
   "retention": "zero-retention",
   "notes": "Meituan. 1M context with image input, and a reasoning toggle rather than a tier, so no variant is written. It read a screenshot correctly in a live call, which is what makes it the zero-retention fallback for the vision lane. Measured at 9.4-13.2s. A preview endpoint: it can be withdrawn without a rename.",
   "limits": {
@@ -14,7 +14,7 @@
   },
   "supports": {
     "reasoning": true,
-    "temperature": true,
+    "temperature": false,
     "toolCall": true,
     "structuredOutput": false,
     "attachment": true

package/profiles/minimax-m3.1-flash-free.json

@@ -5,7 +5,7 @@
   "provider": "opencode",
   "model": "minimax-m3.1-flash-free",
   "family": "minimax",
-  "verified": "2026-10-19",
+  "verified": "2026-10-21",
   "retention": "zero-retention",
   "notes": "Flagship coding and agent model on the free tier: 512k output tokens, the widest of any free route, and the fastest measured image reader at 7.7s. Tool calls are stable. It does not support temperature, and models.dev publishes no structured_output flag for it, so JSON output is requested in the prompt rather than negotiated with the API. Measured at 3.1s for a one-word reply.",
   "limits": {
@@ -14,7 +14,7 @@
   },
   "supports": {
     "reasoning": true,
-    "temperature": false,
+    "temperature": true,
     "toolCall": true,
     "structuredOutput": false,
     "attachment": true