agentleFS
iktok90-designiktok90-design/ai-smart-contract-auditor/skills/false-positive-feedback-loop/SKILL.md

false-positive-feedback-loop

Skill36 starsChanged 5 months agoMITView on GitHub

Meta-skill for managing user-dismissed findings. Before reporting any finding, check it against the project's .auditsentry.yml ignore list and inline auditsentry-ignore markers. Activate on every audit command.

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 2 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    If F's fingerprint matches any entry in .auditsentry.yml ignore: → check freshness.

    line 48
  2. Very specific:

    If the surrounding lines (±3 of line) have changed since dismissed_at (use git blame or content hash) → fingerprint invalidates, re-surface …

    line 49
  3. Distinctive:

    If matches F's ID or vuln class → skip F.

    line 54
  4. Very specific:

    If F's file path matches .auditsentryignore → skip F.

    line 57
  5. Distinctive:

    If F survived all checks, include in the report.

    line 60
  6. Very specific:

    Every dismissed finding is still queryable via /report --include-dismissed.

    line 69

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

False-positive feedback loop (meta-skill)

False positives are the #1 reason audit tools get abandoned. This skill governs the state-persistent mechanism for "user told me this is fine, don't report it again."

Sources of dismissals

1. .auditsentry.yml ignore: section

ignore:
  - id: REENT-001
    path: "src/Vault.sol"
    line: 142
    fingerprint: "sha256:abc..."
    reason: "Single trusted caller, reviewed 2025-Q4"
    dismissed_at: "2026-05-12"
    dismissed_by: "alice@example.com"

2. Inline markers in code

// auditsentry-ignore: REENT-001 — single trusted caller
function trustedOp() external onlyOwner { ... }

Effective for next ~5 lines.

3. .auditsentryignore (file glob)

# .auditsentryignore
test/**
script/**
mocks/**

Skip these paths entirely.

Procedure (run before reporting findings)

For each candidate finding F:

Step 1 — Fingerprint

Compute sha256(file_path + ":" + line + ":" + vuln_class). This is the unique identifier for "this exact issue at this exact location in this code shape."

Step 2 — Check ignore list

  • If F's fingerprint matches any entry in .auditsentry.yml ignore: → check freshness.
  • If the surrounding lines (±3 of line) have changed since dismissed_at (use git blame or content hash) → fingerprint invalidates, re-surface F with note "previously dismissed but code has changed".
  • Otherwise → skip F, do not report.

Step 3 — Check inline markers

Scan within 5 lines above F's location for // auditsentry-ignore: <id>.

  • If matches F's ID or vuln class → skip F.

Step 4 — Check path globs

If F's file path matches .auditsentryignore → skip F.

Step 5 — Surface, with metadata

If F survived all checks, include in the report. If it was previously dismissed but is now re-surfaced due to code change, note this:

[REENT-001 | High] Reentrancy in withdraw (re-surfaced)
  Previously dismissed: 2025-Q3 (by alice@example.com, reason "single trusted caller")
  Re-surfaced because: code around line 142 changed since dismissal

Audit trail

Every dismissed finding is still queryable via /report --include-dismissed. This:

  • Lets auditors verify what was suppressed.
  • Catches dismissals being abused to hide real bugs.
  • Provides accountability.

Force re-surface

/audit --no-ignore ignores the ignore list. Useful for fresh re-audits or external review.

Constraints

  • Critical findings can't be silently dismissed. /dismiss CRITICAL-xxx requires --force flag and --reason of ≥30 chars.
  • Dismissals expire on code change. Fingerprint binds to code shape.
  • Dismissals require a reason. Silent dismissals are not allowed.

Why this matters

A first-run audit produces 30 findings, of which 5 are real and 25 are false positives. Without a feedback loop, the user dismisses them by not running the tool again. With the loop, the user dismisses the 25 with reasons, and the next run produces 5 — the real ones. Trust compounds.

Related

  • /dismiss (command)
  • [[confidence-scoring]] — LOW-confidence findings should default to "soft-dismiss" with auto-recheck
  • [[multi-pass-self-critique]] — dismissals from prior runs inform future passes

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.