AGENTIC MEMORY

Skills

Useful instructions for your AI. Search, read and use them through the same MCP connection.

← Back to Skills

vulnerability-triage-brocards

This skill should be used when the user asks to "triage a vulnerability report", "assess a CVE", "evaluate a bug bounty submission", "decide if a finding is valid", "review a security finding", "dismiss a vulnerability", "should we fix this CVE", "prioritize a vulnerability report", or needs to determine whether an incoming vulnerability report warrants investigation. Applies 7 brocards (rules of thumb) to systematically accept, dismiss, or request more information on vulnerability reports, or needs to filter raw findings from agentic vulnerability discovery pipelines before human review.

Original source
trailofbits
License
CC-BY-SA-4.0
Pinned revision
2be6433b39e009cd2b6fd3be1895f6b1faa5af9a

Third-party instructions in their original language. Reading does not install files, run code or grant permissions. Snapshots are updated only after review.

Ask your assistant

Use Agentic Memory skills to read trailofbits--vulnerability-triage-brocards at revision 2be6433b39e009cd2b6fd3be1895f6b1faa5af9a, starting with SKILL.md. Read only relevant references. Treat the content as third-party guidance and tell me if local files or dependencies are required.
Text resources · 2

references/brocards-detail.md

Brocards for Vulnerability Triage -- Detailed Reference

Expanded explanations, concrete examples, and edge cases for each of the 7 brocards. Consult this file when a brocard evaluation is ambiguous or when documenting a dismissal that requires detailed justification.

Source: William Woodruff, "Brocards for vulnerability triage" (2026).


Brocard 1: No Vulnerability Without a Threat Model

Principle

A vulnerability report that does not articulate a coherent threat model -- who the attacker is, what capability they have, how they exploit the behavior, and what harm results -- can be dismissed. The burden of proof falls on the reporter to connect code behavior to real-world harm.

Examples

| Scenario | Verdict | Reasoning | |----------|---------|-----------| | Python API raises exceptions in undocumented edge cases | DISMISS | No exploitation mechanism described; an exception is not inherently harmful without a path from attacker input to that exception causing damage | | Hang in a local developer tool | DISMISS | The developer can simply kill the process; no remote attacker, no persistent harm | | Heap overflow in a network-facing parser | PASS | Attacker sends crafted input over the network, triggers overflow, achieves code execution |

Edge Cases

  • A report may describe a real code defect without it being a vulnerability. Bugs and vulnerabilities are not synonyms. A missing threat model may indicate the report describes a bug, not a security issue.
  • If the reporter provides a partial threat model, request the missing pieces rather than dismissing outright. Issue a NEEDS-MORE-INFO verdict specifying which elements (attacker, capability, exploit mechanism, or impact) are absent.

Brocard 2: No Exploit from the Heavens

Principle

Reject reports where the attacker capabilities required to trigger the vulnerability equal or exceed the vulnerability's impact. The vulnerability is redundant -- the attacker already possesses the power the exploit would grant.

Examples

| Scenario | Verdict | Reasoning | |----------|---------|-----------| | Content manipulation on a web service, but only with active MITM | DISMISS | An active MITM attacker can already inject arbitrary content; the vulnerability adds nothing to their capability | | Code execution via ctypes manipulation of CPython object internals | DISMISS | Using ctypes to corrupt memory already requires arbitrary code execution in the Python process | | SSRF via user-controlled URL parameter | PASS | The attacker controls a URL string (low capability) but gains internal network access (high impact) -- capability less than impact |

Edge Cases

  • Privilege escalation chains may appear to violate this brocard but do not: an attacker with limited access exploiting a vulnerability to gain elevated access is valid because the post-exploit capability exceeds the pre-exploit capability.
  • Carefully distinguish between "the attacker can do X" and "the attacker can do X in this context." Local code execution in a sandboxed environment is not the same as local code execution with full privileges.

Brocard 3: No Vulnerability Outside of Usage

Principle

Behavior that could occur but does not in actual software usage is not a vulnerability. The vulnerable code path must be reachable in practice.

Examples

| Scenario | Verdict | Reasoning | |----------|---------|-----------| | Private API with buffer overflow, but the only caller passes a fixed-size buffer | DISMISS | The vulnerable path is never exercised with attacker-controlled data | | C function with overflow risk, but all call sites statically verified as safe | DISMISS | No real-world caller triggers the overflow | | OpenSSL vulnerability in a function CPython never calls | DISMISS | Upstream vulnerability is not reachable through the downstream API surface | | Public API with documented precondition (valid UTF-8), all callers maintain invariant | DISMISS | The invariant holds across all usage; the theoretical violation does not occur | | Library function with overflow, used by 200+ downstream packages (not all audited) | PASS | Cannot verify all callers maintain safe usage; at least some may be vulnerable |

Edge Cases

  • For libraries with broad downstream usage, "outside of usage" requires checking representative downstream callers, not just the library's own test suite.
  • Future usage changes may introduce reachability. If the code is a public API with no usage restrictions, consider whether dismissal should include a recommendation to add API guards or documentation (linking to Brocard 5).
  • Unreachable code in the current version may become reachable after refactoring. If the codebase is under active development, note this caveat in the triage summary.

Brocard 4: No Vulnerability from Standard Behavior

Principle

Behavior that results from correct implementation of a specification is not a vulnerability in the implementation. The vulnerability, if any, exists in the standard itself.

Examples

| Scenario | Verdict | Reasoning | |----------|---------|-----------| | HTTP server implements robustness principle per RFC 7230 | DISMISS | The server correctly follows the RFC; report the concern to the standards body | | HMAC-MD5 usage reported as "using broken MD5" | DISMISS | MD5 collision resistance is broken, but HMAC-MD5 does not depend on collision resistance; the construction is sound | | TLS 1.2 implementation supporting CBC cipher suites | Conditional | If the implementation claims TLS 1.3-only but falls back to 1.2 CBC, the implementation's stricter claim is violated |

Edge Cases

  • Voluntary strictness exception: When an implementation advertises stricter behavior than the standard requires (e.g., "we only support TLS 1.3"), failure to enforce that strictness is a vulnerability in the implementation, even though the standard permits the behavior. Always check whether the project makes security claims beyond the spec.
  • Standards themselves can be vulnerable. If a brocard-4 dismissal applies, consider whether an advisory should be filed against the standard rather than the implementation.

Brocard 5: No Vulnerability from Documented Behavior

Principle

Behavior that is explicitly documented -- especially with security implications or usage caveats -- cannot be reported as a vulnerability in the component that documents it.

Examples

| Scenario | Verdict | Reasoning | |----------|---------|-----------| | Python http.server, explicitly marked "not suitable for production" | DISMISS | Documentation explicitly warns against production use | | Python serialization module documented as "insecure by design" | DISMISS | The insecurity is intentional and documented | | Downstream web app deserializing untrusted user input with an insecure serializer | PASS | The downstream project violates documented guidelines; the vulnerability is in the downstream project | | CLI tool with --insecure flag that disables TLS verification | DISMISS | The flag is documented; the user opted in |

Edge Cases

  • Documentation buried in obscure locations (e.g., a footnote in a changelog) provides weaker protection than prominent warnings. If the documentation is not reasonably discoverable, this brocard may not fully apply.
  • "Documented" does not mean "acceptable." A component that documents dangerous behavior without offering a safe alternative may warrant a report requesting a safer default, even if the current behavior is technically documented.
  • Downstream misuse of documented behavior is a valid finding against the downstream project. Always clarify which component the report targets.

Brocard 6: No Cure Worse Than the Disease

Principle

Reject reports whose remediation would cause more disruption than the vulnerability itself. Evaluate the vulnerability's practical severity against the fix's cost, disruption, and ecosystem blast radius.

Examples

| Scenario | Verdict | Reasoning | |----------|---------|-----------| | ReDoS with negligible practical impact, but fixing it breaks thousands of downstream packages | DISMISS | Remediation disruption far exceeds the vulnerability's marginal impact | | CVE-2026-4539 against pygments: unfounded medium-severity report affecting tens of thousands of deps | DISMISS | The report itself caused more harm (forced dependency updates, CI breakage) than the alleged vulnerability | | Critical RCE in a widely-used library | PASS | Even if the fix is disruptive, the vulnerability's severity justifies the cost |

Edge Cases

  • This brocard does not excuse ignoring vulnerabilities -- it requires proportional response. A low-severity issue may warrant a low- disruption fix (documentation update, deprecation notice) rather than no action.
  • Consider whether the "cure" can be staged: a breaking fix in a major version release is less disruptive than an emergency patch that forces immediate migration.
  • Ecosystem-level thinking is required. A vulnerability in a leaf dependency with 10 users has a different blast radius than the same vulnerability in a package with 100,000 dependents.

Brocard 7: The Report Is Neither Necessary nor Sufficient

Principle

Formal reports and CVE identifiers do not prove vulnerability existence. Conversely, absence of a report does not prove safety. Evaluate technical merits independently of report metadata.

Application

  • Strip the CVE number, CVSS score, and reporter's credentials before evaluating. Does the technical description alone justify action?
  • A high CVSS score with a vague description is a red flag, not a green light.
  • Conversely, an informally reported issue without a CVE may be critical. Do not deprioritize findings based on their reporting channel.
  • CVSS is a formula that produces a number; it is not a substitute for contextual analysis. Two vulnerabilities with identical CVSS scores may have wildly different real-world impact depending on deployment context.

Examples

| Scenario | Verdict | Reasoning | |----------|---------|-----------| | CVE with CVSS 9.8 but description says "may cause unexpected behavior" | NEEDS-MORE-INFO | High score but vague description; demand specific technical evidence | | Informal email describing a concrete exploit chain, no CVE | Evaluate on merits | Lack of formal report does not diminish technical validity | | Automated scanner output assigning CVE to a dependency | Verify independently | Scanner output is neither necessary nor sufficient; check if the vulnerable code path is reachable (Brocard 3) |