refactor(traces): remove local fixed analyst stack

[drewstone]

Deletes the unused local five-analyst menu, fixed prompts, parser, router call, tests, and generated API reference: 581 lines removed. Runtime must use the canonical agent-eval registry over captured tool spans instead of maintaining a weaker duplicate.

Checks:

• lint: 500 files passed
• typecheck: passed, including examples
• build: passed
• docs regeneration/freshness: passed
• tests: 1,874 passed, 6 skipped across 174 files (GIT_ALLOW_TEST_IDENTITY=1 for fixture repositories)
• merge-tree with origin/main: clean

[tangletools]

✅ Auto-approved drewstone PR — fd4f3d74

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 · 2026-07-30T09:48:02Z}

[tangletools]

🟡 Value Audit — sound-with-nits

Verdict	sound-with-nits
Concerns	2 (2 weak-concern)
Heuristic	0.0s
Duplication	0.0s
Interrogation	111.1s (2 bridge agents)
Total	111.1s

💰 Value — sound-with-nits

Deletes a genuinely-unused, self-admitted weaker duplicate of the canonical agent-eval analyst registry; correct in-grain cleanup, with one stale doc pointer left behind.

• What it does: Removes src/mcp/tools/checks.ts (a standalone 'lens directory' of 5 fixed analyst kinds + its own LLM router caller, raw-row validator, finding schema prompt, and run/dispatch runner), its test (tests/kernel/checks.test.ts), the public re-exports from src/mcp/index.ts, and the generated API reference for those symbols (docs/api/mcp.md, docs/api/primitive-catalog.md). Net -581 lines, no behavior ch
• Goals it achieves: Kill an unused parallel analyst implementation so there is one canonical lens path instead of two. The deleted file's own header concedes the goal: it called itself 'a deliberate SUBSET of agent-eval's full TraceAnalystKindSpec' and a kind that 'needs the heavy agentic actor upgrades to createTraceAnalystKind.' Keeping it meant two places to maintain the finding shape, the lens taxonomy, and the L
• Assessment: Good change, in the grain of the codebase. I confirmed via grep that the deleted symbols (Check, CheckRunnerOptions, defaultChecks, liftFindings, makeCheckRunner, renderTrace, runCheck) have ZERO callers anywhere in src/ or tests/ after this PR — the only references were the deleted re-export and the deleted test. The canonical replacement is real and actively wired: (1) AnalystRegistry in src/mcp
• Better / existing approach: none — this is the right approach. Searched src/ and tests/ for every deleted symbol and for any import of './tools/checks' / '@tangle-network/agent-runtime/mcp' that pulled these names: no internal consumer exists. The canonical registry the PR names is present and wired, so deletion rather than extension/consolidation is correct. A facade/re-export would just re-add the surface-area debt the PR
• Model: opencode/zai-coding-plan/glm-5.2
• Bridge attempts: 2
• Bridge warning: opencode/kimi-for-coding/k2p7: bridge stream ended without value-audit content

🎯 Usefulness — sound-with-nits

Clean deletion of a weaker duplicate analyst stack; the canonical agent-eval registry path that already feeds the live coordination/analyst-loop surface is untouched and remains the sole lens directory.

• Integration: Reachable and correct. Grep across src/tests/bench/examples for the removed symbols (defaultChecks, makeCheckRunner, liftFindings, runCheck, renderTrace, Check, CheckRunnerOptions, 'tools/checks') returns zero consumers after removal. The in-process seam (AnalystRegistry at src/mcp/tools/coordination.ts:71, still exported from src/mcp/index.ts:152) and the eval seam (TraceAnalystKindSpec at src/ag
• Fit with existing patterns: Fits the codebase grain and removes a competing pattern. The deleted module's own header (checks.ts:1-13) stated a Check is 'a deliberate SUBSET of agent-eval's full TraceAnalystKindSpec' and that a kind needing real agentic behavior 'upgrades to createTraceAnalystKind' — i.e., the codebase already had the stronger canonical factory. Keeping both meant two finding-shape validators, two runners (th
• Real-world viability: Deletion, so nothing new to hold up; the canonical path carries the load. The removed makeCheckRunner returned Promise<AnalystFinding[] | {error}> while the live AnalystRegistry.run returns Promise — they were never interchangeable, confirming the removed one was a parallel, not a backing, implementation. No concurrency/error-path surface is introduced or removed by this change.
• Model: opencode/zai-coding-plan/glm-5.2
• Bridge attempts: 1

💰 Value Audit

🟡 Stale pointer to the deleted file left in docs/architecture.md [maintenance] ``

docs/architecture.md:520 still cites 'src/mcp/tools/checks.ts:93 — defaultChecks' as the canonical lens menu and §13.5 describes analysts as inline lenses with 'harness null' reading from that file. This PR regenerated docs/api/mcp.md and docs/api/primitive-catalog.md (the API refs) but did not touch the hand-written architecture doc, so it now points at a deleted file. The PR title says it removes 'the fixed analyst reference' — update docs/architecture.md §13.5 to point at the canonical path (

🎯 Usefulness Audit

🟡 Stale architecture doc still points at the deleted file [integration] ``

docs/architecture.md:520 still cites 'src/mcp/tools/checks.ts:93 — defaultChecks' and claims 'the driver picks lenses via list_analysts/run_analyst', contradicting this PR's body assertion that 'docs regeneration/freshness: passed'. A future integrator following that section lands on a deleted file. Update §13.5 to reference the canonical agent-eval kind directory (or coordination.ts's AnalystRegistry) so the doc matches the post-removal reality. This is a docs-pass item, but it gates the 'docs

---

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 · 20260730T095016Z}

[tangletools]

✅ No Blockers — fd4f3d74

Review health 100/100 · Reviewer score 95/100 · Confidence 70/100 · 0 findings (none)

glm: Correctness 95 · Security 95 · Testing 95 · Architecture 95

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

Full multi-shot audit completed 2/2 planned shots over 3 changed files. Global verifier still owns final merge decision.

No findings.

---

_{tangletools · 2026-07-30T09:51:12Z · trace}

[tangletools]

✅ Clean — fd4f3d74

Full multi-shot audit completed 2/2 planned shots over 3 changed files. Global verifier still owns final merge decision.

Full immutable report for this review: trace

Summary comment for this run: full summary

---

_{tangletools · 2026-07-30T09:51:12Z · immutable trace}