refactor(primeintellect): remove legacy single-turn adapter

[drewstone]

Summary

• remove the legacy single-turn PrimeIntellect export and substring scoring
• remove its package entry, build entry, tests, and README example
• direct users to the full runtime-program adapter in feat(primeintellect): run runtime programs on Verifiers v1 agent-runtime#553

Checks

• pnpm test: 3,217 passed, 2 skipped
• pnpm typecheck
• pnpm lint: exited 0 with 4 existing warnings outside this change
• pnpm build
• pnpm verify:package
• git merge-tree --write-tree origin/main HEAD

[tangletools]

✅ Auto-approved drewstone PR — e6cbc516

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 · 2026-07-15T05:16:22Z}

[tangletools]

🟢 Value Audit — sound

Verdict	sound
Concerns	0 (none)
Heuristic	0.0s
Duplication	0.0s
Interrogation	64.3s (2 bridge agents)
Total	64.3s

💰 Value — sound

Removes a PrimeIntellect Verifiers environment exporter that was added hours earlier, relocating runtime-execution surface to agent-runtime where it belongs per the repo's layering rule — clean, complete removal with a CHANGELOG migration pointer.

• What it does: Deletes src/primeintellect/index.ts (464 LOC) which exported RunRecords+scenarios into a PrimeIntellect Verifiers Python package (pyproject.toml, dataset.jsonl, run_records.jsonl, and a generated load_environment() using vf.SingleTurnEnv with substring scoring). Also deletes its 146-line test, the ./primeintellect package.json subpath entry, its scripts/build-entries.mjs line, and the READ
• Goals it achieves: 1) Respect the substrate/runtime layering rule that CLAUDE.md states explicitly — agent-eval must hold only primitives that 'make sense WITHOUT a running agent loop'. Generating and executing a Python vf.SingleTurnEnv is runtime-shaped, not substrate-shaped. 2) Avoid keeping a 'legacy single-turn adapter' whose substring-based reward (required_substrings → fraction-of-hits) is a weaker evalu
• Assessment: Good. The removal is complete and coherent: code, tests, package export, build entry, README section all removed; CHANGELOG marks it breaking with a concrete migration path; repo has zero remaining references to the removed symbol (only an unrelated renderers link in examples/publish-rl-dataset/README.md and a comment in src/rl/dataset.ts:247 remain, both pointing at a different PrimeIntellect p
• Better / existing approach: none — this is the right approach. Searched src/rl/ (exporters.ts, run-record-adapters.ts, dataset.ts, corpus.ts) and the wider tree for any existing Verifiers-environment / SingleTurnEnv / substring-scoring capability — none exists; the RL module handles generic SFT/RL dataset export, not Python Verifiers packaging. The replacement lives in agent-runtime#553 per the PR body, which matches the l
• Model: opencode/kimi-for-coding/k2p7
• Bridge attempts: 1

🎯 Usefulness — sound

Clean removal of a legacy single-turn PrimeIntellect export that used unsafe substring scoring, replaced by a full runtime-program adapter in the correct layer (agent-runtime); nothing in this repo still references it.

• Integration: Deletion is fully consistent — package.json subpath, build-entries.mjs entry, source file, test file, and README example all removed together; grep across *.ts and *.md finds zero remaining imports of the deleted module. The CHANGELOG (line 11) records the breaking change and points users to @tangle-network/agent-runtime/primeintellect. The replacement lives in a sibling package per tangle-network
• Fit with existing patterns: Matches the documented layering rule in CLAUDE.md: agent-eval is the substrate with zero upward deps on agent-runtime, and a Verifiers adapter that needs real runtime execution + full trace import belongs in agent-runtime, not here. Moving the real implementation to the runtime package and stripping the unsafe single-turn shim from the substrate is exactly the grain of this codebase.
• Real-world viability: The removed module scored via substring containment, which the CHANGELOG explicitly calls out as unsafe; the replacement does runtime execution with full trace import, so the deletion removes a known-weak scoring path rather than a working one. No error/edge-case concerns apply to a pure deletion with no remaining call sites.
• 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 · 20260715T052053Z}

[tangletools]

✅ No Blockers — e6cbc516

Review health 100/100 · Reviewer score 92/100 · Confidence 80/100 · 1 finding (1 low)

	glm	kimi-code	aggregate
Readiness	95	92	92
Confidence	80	80	80
Correctness	95	92	92
Security	95	92	92
Testing	95	92	92
Architecture	95	92	92

Reviewer score is advisory once the run is complete and the verdict has no blockers.

Full multi-shot audit completed 4/4 planned shots over 4 changed files. Global verifier still owns final merge decision. | Full multi-shot audit completed 4/4 planned shots over 4 changed files. Global verifier still owns final merge decision.

🟡 LOW Breaking subpath removal will hard-fail unmigrated consumers — package.json

Deleting the './primeintellect' export means any downstream consumer still importing '@tangle-network/agent-eval/primeintellect' gets ERR_PACKAGE_PATH_NOT_EXPORTED at import time (not a deprecation warning). This is intentional and documented as Breaking in CHANGELOG.md with a migration path to @tangle-network/agent-runtime/primeintellect, and acceptable under 0.x semver, but the blast radius depends on fleet consumers being migrated before this is published. No code change needed in this file; just confirm consumers are off the subpath before release.

---

_{tangletools · 2026-07-15T05:25:17Z · trace}

[tangletools]

✅ Approved — 1 non-blocking finding — e6cbc516

Full multi-shot audit completed 4/4 planned shots over 4 changed files. Global verifier still owns final merge decision. | Full multi-shot audit completed 4/4 planned shots over 4 changed files. Global verifier still owns final merge decision.

Full immutable report for this review: trace

Summary comment for this run: full summary

---

_{tangletools · 2026-07-15T05:25:17Z · immutable trace}