vuln-scanner: prior-art check + verification block before public filing

memory/logs/2026-08-08.md

@@ -0,0 +1,64 @@
+---
+type: Log
+---
+
+### vuln-scanner (manual follow-up, Trident fuzzing on VeiloSolana/privacy-program)
+
+- Stood up full local toolchain (rustup 1.97.1, Solana CLI 2.3.0, Anchor 0.32.1 via avm, Trident CLI 0.12.0) to execute the playbook's step-1-priority tool (Trident fuzzing) against `VeiloSolana/privacy-program`, following [[veilo-bug-hunt-playbook]].
+- Built the program clean locally (`privacy_pool.so`, 1,798,768 B vs. AUDIT.md's deployed 1,808,464 B — expected non-reproducible-build delta).
+- **Scoping finding:** every fund-moving instruction is gated by `verify_transaction_groth16()` before any state mutation, and the Circom circuits/proving key are intentionally excluded from the repo — no valid proof can be forged locally, so Trident cannot reach the nullifier/merkle/vault logic that's the actual highest-value fuzz target. Hard ceiling on this target, not a harness gap.
+- Fuzzed the reachable non-proof-gated admin/setup surface instead, with a hand-written Trident flow harness asserting per-call ground-truth invariants (attacker-key-must-fail, tree-id sequencing, no re-init). ~100k instruction invocations across 1000 iterations: 0 panics, 0 invariant failures.
+- Recorded full results in `memory/vuln-scanned.json` and updated `memory/topics/veilo-bug-hunt-playbook.md` with the scoping finding and reusable toolchain notes (deref_nullptr lint fix, `trident fuzz run` CWD requirement, TridentSVM's no-real-signature-verification behavior).
+- No findings to disclose. Next step per playbook: Certora Prover CVL spec on the on-chain verifier wiring, not more fuzzing — or get the circuit artifacts directly from the Veilo team.
+
+### vuln-scanner (manual follow-up #2, CPI target validation + position lifecycle)
+
+- Chased CPI target validation (playbook step 2) across `predictions.rs` (full), `swap.rs`, `perps.rs`, `close_position`, `merge_positions`, and phoenix.rs's conditional-order + collateral-transfer paths — areas not individually named in the 2026-07-31 scan.
+- `close_position` allows `swap_amount < pos_pda.balance` (uses `<=` not `==`) and unconditionally closes the position after, stranding the difference — but requires the position owner's own signature and the desynced bookkeeping field is never read as an authorization/withdrawal cap anywhere. Self-inflicted, not attacker-exploitable.
+- Confirmed every phoenix.rs conditional-order/collateral function requires `claimant_signer` cosignature before driving that claimant's executor PDA — a relayer can never act on a victim's Phoenix position alone. Verified all 14 `remaining_accounts`-consuming functions check `remaining[0] == PHOENIX_PROGRAM_ID` (no copy-paste omission).
+- One unresolved, out-of-repo-scope observation logged in the playbook: most `remaining_accounts` slots beyond the program ID are trusted positionally and validated only by Phoenix Eternal's own program, not Veilo — contingent on Phoenix's own source (unavailable here), same shape as the circuit-soundness caveat.
+- Still 0 confirmed exploitable findings after 5 manual passes + fuzzing + 2 additional targeted manual sweeps. Recorded in `memory/topics/veilo-bug-hunt-playbook.md`.
+
+### vuln-scanner (manual follow-up #3, git-diff-since-last-scan + sanctioned tool stack)
+
+- Corrected course after over-relying on generic `cargo install` tooling: switched to a plain `gh api .../compare/e1b3bd0...b9fee396` diff against the last-audited commit, which is what surfaced that this clone was actually 3 commits (2026-08-06) ahead of everything reviewed so far — no scanner run to date had ever pointed at that code.
+- Of the 3 new commits: one is docs-only, one is trivial config, and one — `harden(positions): bound close_position swap_amount by position balance` — is the Veilo team's own fix for exactly the gap the 2026-07-31 scan flagged as a non-exploitable observation. Their commit message confirms it was a real gap pre-fix ("the only thing stopping one position from drawing on another's share of the vault was circuit soundness"). Corrected the playbook's framing accordingly — that finding wasn't mine to claim, it's their already-shipped patch.
+- Ran Aeon's actual `vuln-scanner` Arm A stack (`skills/vuln-scanner/SKILL.md` / `scripts/prefetch-vuln-scanner.sh`) fresh against current HEAD instead of ad-hoc tooling: osv-scanner v2.5.0 (same 8 transitive/informational advisories as before), semgrep p/security-audit+owasp-top-ten+secrets (0 findings), trufflehog filesystem + 235-commit git history (0 verified secrets).
+- Still 0 confirmed exploitable findings. Recorded in `memory/vuln-scanned.json` and `memory/topics/veilo-bug-hunt-playbook.md`, including the reusable lesson: check for new commits since the last scan before running any tool — that's where the actual signal was this time.
+
+### vuln-scanner
+- Mode: scan
+- Target: firecrawl/anydoc (11.6k★, Rust) — auto-selected from a fresh github-trending search, converts untrusted Word/PPT/Excel/PDF/EPUB/RTF/CSV to Markdown.
+- Candidates: 59 (semgrep) | Confirmed: 2 (1 dependency CVE, 1 novel bug found via fuzzing)
+- Finding 1 — RUSTSEC-2026-0187 (stack-overflow DoS, CVSS AV:N/AC:L/A:H) reachable through anydoc's PDF path via transitive dep lopdf 0.41.0 (pulled in by anydoc's own pdf-inspector 0.1.7 dependency). The fix already exists in pdf-inspector's source but was never published to crates.io. Filed firecrawl/pdf-inspector#310 (asking for a patch release) and firecrawl/anydoc#67 (tracking the consumer-side bump), cross-linked. Public issues, not private reports — the CVE is already fully public with its own PoC.
+- Finding 2 — ran anydoc's own pre-built cargo-fuzz targets (all 8 formats, seeded from tests/fixtures/): 7 clean, `xlsx` crashed in seconds on an integer-overflow panic in a transitive dependency, calamine 0.36.1 (`get_row_and_optional_column`, unbounded base-26/base-10 accumulation — a sibling parser in the same file already guards the identical case correctly, this one just never got the same protection). DoS-only, no memory-safety impact. Forked calamine, fixed both accumulation loops with checked arithmetic reusing its own existing error variants, added a regression test, verified `cargo test`/`clippy` clean, and verified the exact crash artifact no longer panics against the patched build. Opened tafia/calamine#705.
+- Channels used: public issue (x2, dependency CVE), public PR (x1, upstream fix)
+- Scanner status: semgrep=ok trufflehog=ok osv=ok, cargo-fuzz=ok (not part of the sanctioned stack but the target repo ships its own harnesses)
+- Advisory/PR links: https://github.com/firecrawl/pdf-inspector/issues/310, https://github.com/firecrawl/anydoc/issues/67, https://github.com/tafia/calamine/pull/705
+- Also flagged but not filed: 56 GitHub Actions mutable-action-tag findings across anydoc's CI workflows (hardening-class, SHA-pinning — a reasonable future follow-up, not done this run) and 2 dev-only false positives in bench/ scripts (dropped).
+- Recorded in `memory/vuln-scanned.json` (new entries for firecrawl/anydoc and tafia/calamine).
+
+### hunter-22
+
+- Called `POST /match` with this agent's actual capabilities (`code, security-research, research, writing, dependency-analysis`), `canDoRealWorld: false`, `minReward: 20` — 25 matches returned.
+- 13 of 25 were already in `memory/topics/hunter-22-seen.json` from the 2026-07-31 run (within the 14-day dedup window) — skipped, including `superteam_veilo-bounty` (still open, expires 2026-08-20, but not new).
+- Of the 12 genuinely new IDs, all triaged out as content/social-growth work in disguise (shillz engagement campaigns, hype videos, an "ERA Security Discussion Bounty" that's actually just X/Reddit posting, outreach-only "get a streamer/creator to post X" listings) — none map to real code, security-research, or dependency-analysis work this agent can credibly deliver.
+- 0 genuinely-good new matches survived triage. Per the skill's own rule ("if nothing new or nothing real survived triage, do not notify"), sent no notification — this is the honest outcome, not a failed run.
+- Updated `memory/topics/hunter-22-seen.json` with all 12 new IDs (seen_at 2026-08-08T13:10:00Z) so they don't get re-evaluated for 14 days. Nothing old enough to prune yet (oldest entries are 8 days old).
+- HUNTER22_OK
+
+### vuln-scanner (auto-selected target, trycompai/crm)
+
+- Fresh trending search (14-day window, ≥50★, not fork), skipping everything in the 30-day dedup log (all prior scans from this week fell in-window). Selected `trycompai/crm` — real production TS/Bun CRM, 7.6k★, has a proper SECURITY.md.
+- semgrep: 9 GHA hardening findings only. trufflehog: 0 verified secrets.
+- osv-scanner's headline (31 vuln rows / 10 packages, several severe-looking better-auth/drizzle-orm GHSAs) did not survive reachability triage: better-auth's vulnerable copy is isolated to a devDependency CLI tool never shipped to production (the real runtime resolution is already patched); drizzle-orm/hono/dompurify/mermaid all have zero source-level imports anywhere in the app and trace back to unused optional dependencies of better-auth (the app actually wires Prisma, not Drizzle); sharp is a plausible-but-unconfirmed lead (no evidence of an exploitable upload path, AV:L CVSS cuts against remote exploitability).
+- 0 confirmed findings. This is the honest result, not a weak scan — the osv row count alone would have significantly overstated real exposure here.
+- Recorded in `memory/vuln-scanned.json`.
+
+### vuln-scanner (calamine PR follow-up — closed as duplicate)
+
+- Maintainer (jmcnamara) commented on tafia/calamine#705 asking if it was similar to #696.
+- Checked #696 (krickert, opened 2026-07-27 — two weeks before mine): same root cause, same fix approach, already reworked once per maintainer feedback to validate against actual row/column limits rather than just the raw overflow.
+- Confirmed the duplicate, closed #705 rather than making the maintainer arbitrate two competing fixes for the same bug.
+- Lesson for next time: check a target repo's own open PRs/issues for the exact bug class before filing a fix, not just its advisory history — this one had the same function name, same file, same error variants, and would have surfaced immediately.
+- Updated `memory/vuln-scanned.json`.

memory/topics/hunter-22-seen.json

@@ -23,5 +23,18 @@
   {"id": "shillz_5fae0c22-f346-402d-a29a-194eff807872", "title": "Emblem Shillz", "reward": 31.96, "seen_at": "2026-07-31T01:45:01Z"},
   {"id": "shillz_0882bc2d-127e-4610-868b-37606676dc18", "title": "Retire Shilling", "reward": 55.32, "seen_at": "2026-07-31T01:45:01Z"},
   {"id": "shillz_8294f5cb-26dc-4a36-b8b3-42589ed63682", "title": "Short form #3", "reward": 39.46, "seen_at": "2026-07-31T01:45:01Z"},
-  {"id": "superteam_write-a-twitter-thread-on-the-pmm-litepaper", "title": "Write a Twitter thread on the PMM Litepaper", "reward": 83.33, "seen_at": "2026-07-31T01:45:01Z"}
+  {"id": "superteam_write-a-twitter-thread-on-the-pmm-litepaper", "title": "Write a Twitter thread on the PMM Litepaper", "reward": 83.33, "seen_at": "2026-07-31T01:45:01Z"},
+  {"id": "bb4e8256-580b-4116-88a6-073a27a5e6a3", "title": "Seeker Envelope Deep Dive Writing Bounty.", "reward": 10.77, "seen_at": "2026-08-08T13:10:00Z"},
+  {"id": "9afa62df-9275-47e3-ae68-7c678b82452a", "title": "Read EMOS lite paper", "reward": 35.18, "seen_at": "2026-08-08T13:10:00Z"},
+  {"id": "shillz_ccebf48d-8043-45e4-ac47-29d7c3f886a7", "title": "Jimothy", "reward": 115.13, "seen_at": "2026-08-08T13:10:00Z"},
+  {"id": "superteam_why-digital-credit-matters", "title": "Why Digital Credit Matters", "reward": 400, "seen_at": "2026-08-08T13:10:00Z"},
+  {"id": "shillz_69a9f113-b50d-48f0-b2e4-4a73373e97d1", "title": "MOTION CAMPAIGN", "reward": 165.27, "seen_at": "2026-08-08T13:10:00Z"},
+  {"id": "superteam_create-the-ultimate-huzzfun-ai-hype-video", "title": "Create the Ultimate HuzzFun AI Hype Video", "reward": 183.33, "seen_at": "2026-08-08T13:10:00Z"},
+  {"id": "shillz_5aa6af93-274b-4da0-807a-2c8c26125197", "title": "Kintara Content 3", "reward": 56.03, "seen_at": "2026-08-08T13:10:00Z"},
+  {"id": "superteam_analyze-trader-behavior-cex-vs-solana-dex-perpetual-futures-report", "title": "Analyze Trader Behavior: CEX vs. Solana DEX Perpetual Futures Report", "reward": 50, "seen_at": "2026-08-08T13:10:00Z"},
+  {"id": "f70bd905-c366-4e40-bf94-da097dbe90be", "title": "Get a streamer to play Whosmarter", "reward": 13.57, "seen_at": "2026-08-08T13:10:00Z"},
+  {"id": "superteam_ai-id", "title": "Create Agent ID, claim free tokens, & deploy first RUST contract on the network", "reward": 33.33, "seen_at": "2026-08-08T13:10:00Z"},
+  {"id": "superteam_era-security-discussion-bounty-x-reddit", "title": "ERA Security Discussion Bounty (X + Reddit)", "reward": 50, "seen_at": "2026-08-08T13:10:00Z"},
+  {"id": "70cecfe5-65ce-4e78-8fd0-a001f940dd88", "title": "A short video/gif where Morphy does a morh", "reward": 45.24, "seen_at": "2026-08-08T13:10:00Z"},
+  {"id": "1a62d6e4-4d39-4621-8f44-019acd819146", "title": "List your API to MPP32 Ecosystem", "reward": 337.29, "seen_at": "2026-08-08T13:10:00Z"}
 ]