agentleFS
ToddGBensonToddGBenson/keel/.claude/skills/scanner-triage/SKILL.md

scanner-triage

Skill0 starsChanged 3 months agoApache-2.0View on GitHub

Turn raw SAST, SCA, IaC, secret, and container scanner output into dispositioned findings. Use when reviewing scan results, deciding whether a result is a true positive, assigning severity, writing or auditing a suppression, or when a pipeline is failing on scanner noise. Covers reachability analysis and severity-from-context.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Every result gets exactly one disposition.

    line 17
  2. Distinctive:

    When facing hundreds of results:

    line 113
  3. Distinctive:

    Never mass-suppress to get green: If a group is genuinely all false positives, fix the scanner configuration.

    line 121

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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

Scanner triage

Consumers: security-engineer, /security-gate, process/gates/g4-verified.md, .github/workflows/security.yml.

Scanner output is not findings

A scanner emits results. A finding is a result that has been shown to be real, reachable, and consequential here. Handing a team 400 raw results is not security work; it is transferring the actual work while appearing to have done it.

Every result gets exactly one disposition.

result
  │
  ├─ Is the pattern actually present?  ── no ──▶ FALSE POSITIVE
  │        yes                                    (rationale + expiry required)
  ├─ Is the code reachable?            ── no ──▶ FALSE POSITIVE (unreachable)
  │        yes                                    or NOT EXPLOITABLE — say which
  ├─ Is it exploitable in our context? ── no ──▶ ACCEPTED RISK
  │        yes                                    (human approval + POA&M + expiry)
  └──────────────────────────────────────────▶ TRUE POSITIVE
                                                 (severity + SLA + issue)

Reachability — the step most often skipped

A CVE in a dependency you ship is not automatically a finding. Ask, in order:

  1. Is the vulnerable function called? Many CVEs live in one module of a large library. If you import library.parse and the flaw is in library.serve, say so — with the evidence (a call-graph query, a grep, the absence of the import).
  2. Can attacker-controlled input reach it? A deserialization flaw in a code path that only ever handles internally-generated data is a different risk than one on the request boundary.
  3. What preconditions does exploitation need? Authentication? A specific role? Local access? Each precondition that must hold is a genuine reduction — but only if it is enforced, not merely typical.
  4. Is there a compensating control in front of it? A WAF rule is not a fix; a validated allowlist at the boundary might be.

Record the reasoning. A reachability judgment you cannot reconstruct in six months reads as arbitrary to an assessor, and it will be re-litigated every quarter.

Severity from context, not from the base score

The vendor's CVSS base score assumes a generic deployment. You are not generic.

Raises it hereLowers it here
Unauthenticated pathRequires admin privilege already
Internet-facingInternal-only, network-segmented
Handles personal or regulated dataSynthetic or public data
Known exploited (CISA KEV)No public exploit, theoretical
Our own code, not a dependencyVendor patch already staged

Deviating from the vendor score is correct and expected. Record the deviation and the reason. What is not acceptable is silently accepting the base score and calling that triage.

Never inflate severity to force attention. It spends the credibility that makes your blocks work, and the next genuine Critical gets treated like the last inflated one.

SLAs: Critical 7d · High 30d · Medium 90d · Low next cycle. Critical and High block release.

Writing a suppression that survives audit

A suppression is a claim that a control does not apply here. It needs to be defensible a year later by someone who was not present.

# nosec B608 - query is parameterised via SQLAlchemy bindparams below; the scanner
# does not follow the builder. Verified by test_no_raw_interpolation. See #318.
# expires: 2026-12-01

Required, without exception:

  • What the scanner thinks is wrong
  • Why it is not — specific, and pointing at the thing that makes it not wrong
  • An issue reference
  • An expiry date — expires: YYYY-MM-DD

security.yml's suppression audit fails the build without an issue ref and an expiry. That job exists because unjustified suppressions are how a scanner quietly stops being a control — one reasonable-looking # nosec at a time, until the scan is green and meaningless.

Blanket suppressions fail G4. A file-level or rule-level disable is not a triage decision; it is opting out of the control. If a rule genuinely does not apply to this codebase, disable it in the scanner configuration, in a reviewed PR, with the reasoning — visibly, not in a comment nobody reads.

Suppressions are reviewed quarterly. An expired one is a finding.

Fail on new, not on total

This is the design decision that determines whether the pipeline is a control or decoration.

Gating on the absolute finding count in any real codebase means the pipeline is red on day one and stays red. Within two sprints everyone has learned to click through it, and at that point it detects nothing while still showing green on the assessment.

Gate on the delta the PR introduces. Drive the existing backlog down on SLA through the POA&M, where it has owners and dates.

Batch triage without losing rigour

When facing hundreds of results:

  1. Group by rule, not by file. One rule usually produces one kind of result; judge the rule's applicability once, then confirm the outliers.
  2. Triage the highest severity first, but do not stop there — a cluster of Mediums in one component is often a design problem worth more than a lone High.
  3. Sample within a group and check whether the sample generalises before dispositioning the group. Say in the record that you sampled, and how many.
  4. Never mass-suppress to get green. If a group is genuinely all false positives, fix the scanner configuration.

Writing the finding

So a developer can act without asking you anything:

HIGH — drafts.py:88 — non-transactional quota check The count is read and the write performed outside a transaction. Two concurrent autosaves at quota−1 both pass, exceeding the 5 MB limit. Reachable: any authenticated user, no special preconditions. Reproduce: parallel POST /drafts with the user at 4 of 5 drafts. Fix: take a row lock on the user quota row, or enforce with a DB constraint. SLA: 30 days.

A finding a developer cannot act on is a complaint.

Controls

RA-5 (vulnerability monitoring) · RA-7 (risk response) · SI-2 (flaw remediation) · SA-11(1) (static analysis) · CA-5 (POA&M) · SR-3 (supply chain).

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.