chore(release): align interaction binding cohort

[drewstone]

What changed

• publish Knowledge 7.1.2 against Eval 0.144.6 and Interface 0.46.1
• resolve one installed copy of Core 0.5.4 and Interface 0.46.1
• update the install example and release notes

Proof

• source and contract typechecks passed
• Biome checked 207 files
• 60 test files passed, 3 skipped; 558 tests passed, 12 skipped
• build and packed-package checks passed
• packed consumer resolved one copy each of Eval 0.144.6, Core 0.5.4, and Interface 0.46.1
• official optimizer integration passed 2/2 against published agent-eval-rpc 0.144.6

Release

After merge and green main CI, tag the exact merge commit as v7.1.2.

[tangletools]

✅ Auto-approved drewstone PR — 3b56d3bf

This PR was opened by the trusted drewstone account.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.

_{tangletools · auto-approval · reason: drewstone_author · 2026-08-09T21:50:11Z}

[tangletools]

🟢 Value Audit — sound

Verdict	sound
Concerns	0 (none)
Heuristic	0.0s
Duplication	0.0s
Interrogation	37.6s (2 bridge agents)
Total	37.6s

💰 Value — sound

Standard patch release: bumps Knowledge to 7.1.2 pinning Eval 0.144.6 + Interface 0.46.1 so consumers resolve one interaction-binding cohort — matches the repo's repeated release pattern exactly.

• What it does: Bumps package version 7.1.1 -> 7.1.2 and pins two exact dependencies: @tangle-network/agent-eval 0.144.5 -> 0.144.6 and @tangle-network/agent-interface 0.46.0 -> 0.46.1. Also updates pnpm-lock.yaml, README install example, and adds the 7.1.2 CHANGELOG entry. No source code changes.
• Goals it achieves: Ensure every Knowledge consumer resolves exactly one installed copy of the shared interaction-binding contract across Eval, Core, and Interface (per CHANGELOG: 'one canonical interaction-binding contract through Eval, Core, and Interface'). This is the standard dependency-alignment release this repo ships repeatedly to avoid duplicate-version installs downstream.
• Assessment: Good change, in the grain of the codebase. The git history shows ~8 prior commits with the identical shape (e.g. 4c29d55 'align knowledge with interface 0.46', 1215ad7 'align prompt-cache SDK cohort', b60ce17 'align knowledge with eval 0.143'), and the CHANGELOG entry follows the exact same one-line format. The repo deliberately uses exact-pinned dependency versions (no ^ on the @tangle-network/*
• Better / existing approach: none — this is the right approach. Confirmed by git log that this exact pattern (package.json + pnpm-lock.yaml + README + CHANGELOG, single 'align ' commit) is the established release convention for this repo. No existing tooling or automation is being bypassed; verify:official-optimizers and verify:package scripts (package.json:74-75) exist precisely to validate these bumps, and the PR bo
• Model: opencode/zai-coding-plan/glm-5.2
• Bridge attempts: 2
• Bridge warning: opencode/kimi-for-coding/k2p7: opencode: opencode error

🎯 Usefulness — sound

Routine version-and-dependency alignment release (7.1.1→7.1.2) that pins agent-eval 0.144.6 and agent-interface 0.46.1, following the repo's established release-cohort cadence.

• Integration: Both bumped dependencies are deeply consumed across the source tree — 97 import sites for @tangle-network/agent-eval and @tangle-network/agent-interface (e.g., src/release.ts:9, src/agent-candidate.ts:4, src/memory/improvement/contracts via kb-improvement/contracts.ts:1,6). The package is published with working main/types/exports/bin entry points (package.json:14-49), so consumers reaching
• Fit with existing patterns: Matches the codebase's established pattern exactly. The git log shows a steady cadence of identical chore(release): align ... commits (4c29d55, 1215ad7, 6dc011b, 140ab0f) and the CHANGELOG entry for 7.1.2 is already written in the same one-paragraph 'Updated X to Y' shape as 7.1.1, 7.0.11, 7.0.10. This is the grain, not a competing pattern.
• Real-world viability: Pure pin tightening with no logic change; runtime behavior is governed by the upstream packages, which are pinned exact (no ^/~ on the Tangle deps at lines 78-79) so consumers resolve one copy each, as the PR body claims. The repo's own verify:official-optimizers and verify:package scripts (package.json:74-75) are the gates that back the release claim. No new error paths or edge inputs introduced
• Model: opencode/zai-coding-plan/glm-5.2
• Bridge attempts: 1

No concerns — sound change, no better or existing approach found. ✅

---

What this audit checks

It judges the change on its merits — not whether it was tasked out in an issue. Unticketed, fast-moving work is fine; the question is whether the change is good and whether a better or existing approach should be used instead.

Pass	What it asks
Heuristic	Vague title? Whitespace-only or cruft-bearing diff? (content signals only)
Duplication	Do added function/class names already exist elsewhere in the repo?
Value Audit	What does it do? What goal does it achieve? Is it good? Better architecture or already-exists?
Usefulness Audit	Does it integrate and fit? Will it hold up in real use and actually get used?

Findings are concerns, not blocks — the human reviewer decides what to do with them.

_{value-audit · 20260809T215243Z}