feat(sn60): hot-function triage miner (jimcody1995-20260709-01)

[jimcody1995]

Summary

jimcody1995-20260709-01 updated for July 10, 2026 validation rules — optimized for replica-stable scoring and smoke-test reliability.

What changed (July 2026 rules)

Rule	Our adaptation
2/3 replica pass per project	Depth-first audits on top 1 then top 2–3 graph-ranked files (not wide shallow batches)
Ranking: pass score → passed projects → TPs → invalid runs → precision → F1	Crash-safe agent_main, call-budget tracker (max 3), graceful 429 handling
150k input tokens / problem	Smaller graph digest (14k), 46k deep prompts, 3.2k import context
Smoke test before rounds	Never raises; always returns {"vulnerabilities": [...]}
Helpers allowed	Refactored into helpers/discovery.py, helpers/llm.py, helpers/findings.py
Relay strips sampling knobs	Removed reasoning override; relay pins qwen/qwen3.6-35b-a3b

Strategy (3 calls)

1. Call 1 — graph triage: import-centrality ranked map → target_files only (stable routing)
2. Call 2 — depth-first audit: feat: submit garis agent #1 triage target + import neighbors at full depth
3. Call 3 — depth-first audit: feat: garis candidate agent for taopedia-article repo #2–3 targets at full depth

Self-check

• uv run kata submission validate — pass (screening: pass)
• agent_main() smoke-safe (no crash, valid JSON shape)
• Not AST-equivalent to king (similarity ~0.13)
• Exactly 3 inference calls tracked via CallBudget
• Await next round under new replica scoring

[carlos4s]

Review re-run by carlos4s.

Screening passed. I marked this PR as kata:pending.

[carlos4s]

I closed this PR because it failed the round-start execution screener.

Your PR passed intake earlier, but before scoring Kata runs the candidate once on a real benchmark project to make sure it can execute and return valid report JSON.

Screener project: sherlock_oku_2024_12

Reason:

• SN60 screening execution did not complete successfully: Bitsec execution command timed out after 180.0 seconds.
• SN60 screening report must be a JSON object.

This candidate did not enter the main round scoring, so this is marked kata:invalid rather than kata:losing.

[carlos4s]

I closed this PR because your agent did not beat the current king this round. Improve it and open a new PR to try again.

How your agent compared to the king this round:

• Detection score: 0% (you) vs 4% (king)
• True positives: 0/0 (you) vs 5/117 (king)
• Precision: 0% (you) vs 7% (king)