sast-final-report
Skill2 starsChanged 4 months agoMITView on GitHub
Run LAST, after all per-class SAST detection skills have produced their findings/*.md files. Aggregates every finding, de-duplicates, scores severity, and — the core purpose — reasons about attack chaining: which individual weaknesses combine into full kill chains. Writes SECURITY-ASSESSMENT-REPORT.md.
In this repository49 other files it gives its agents.
- authentication-flawsSkill
- broken-access-control-idorSkill
- business-logic-flawsSkill
- cicd-and-pipeline-injectionSkill
- code-injectionSkill
- command-injectionSkill
- cors-misconfigurationSkill
- crlf-and-header-injectionSkill
- cross-site-scriptingSkill
- csrfSkill
- dependency-and-supply-chainSkill
- dom-clobbering-and-postmessageSkill
- graphql-injection-and-abuseSkill
- hardcoded-secretsSkill
- http-request-smugglingSkill
- iac-misconfigurationSkill
- information-disclosureSkill
- insecure-deserializationSkill
- insecure-randomnessSkill
- jwt-vulnerabilitiesSkill
- ldap-injectionSkill
- llm-and-ai-application-securitySkill
- mass-assignmentSkill
- memory-safetySkill
- mobile-android-securitySkill
- mobile-ios-securitySkill
- nosql-injectionSkill
- oauth-oidc-saml-flawsSkill
- open-redirectSkill
- path-traversal-and-file-inclusionSkill
- prototype-pollutionSkill
- race-conditions-and-toctouSkill
- redosSkill
- sast-finding-validationSkill
- sast-methodologySkill
- security-logging-and-monitoring-failuresSkill
- security-misconfiguration-and-headersSkill
- server-side-template-injectionSkill
- session-managementSkill
- smart-contract-securitySkill
- sql-injectionSkill
- ssrfSkill
- unrestricted-file-uploadSkill
- unsafe-parsers-and-yamlSkill
- weak-cryptographySkill
- web-cache-poisoning-and-deceptionSkill
- websocket-securitySkill
- xpath-and-xml-injectionSkill
- xxeSkill
The file
We split this skill into the rules it gives an agent
10 rules. About 0 are common, 4 are distinctive, and 6 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
6
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 24
Every file in findings/*.md (one per vulnerability class).
- Very specific: line 30
If findings/ is empty or missing, stop and report that no detection skills have run yet — do not fabricate findings.
- Very specific: line 36
Read all findings/*.md. Parse each finding block (template fields).
- Very specific: line 47
Keep Confidence orthogonal — never downgrade severity for uncertainty.
- Very specific: line 49
Read each finding's Validation verdict (set by sast-finding-validation):
- Very specific: line 55
If findings/VALIDATION-LOG.md exists, cite it.
- Distinctive: line 84
Write to the repo root using this structure:
- Distinctive: line 132
Honesty of coverage: If a class skill did not run, say so in §6.
- Distinctive: line 134
Every chain cites real finding IDs: No hypothetical chains without at least one confirmed anchor link.
- Distinctive: line 137
English-only: No Turkish anywhere in the report.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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.
Files
SAST Final Report & Attack-Chaining
Run this only after the detection skills have run — and only when the user asks for a report and/or chained vulnerabilities (it is on-demand, not automatic). It does not hunt for new bugs; it synthesizes what the per-class skills wrote into a single decision-grade report whose centerpiece is how findings chain together.
The chain section is Stage 1 of chaining: it proposes candidate chains as
hypotheses and presents them for the user to choose. Proving them is Stage 2
(sast-finding-validation Phase 3). If the user only wants the chain ideas, you can
emit just the chains section (e.g. CHAINS.md) without the full report.
Inputs
- Every file in
findings/*.md(one per vulnerability class). references/attack-chaining-catalog.md(primitive graph + named chains).references/severity-model.md(per-link and chain severity + chain uplift).findings/VALIDATION-LOG.mdif it exists (per-finding and per-chain verdicts fromsast-finding-validation).
If findings/ is empty or missing, stop and report that no detection skills have
run yet — do not fabricate findings.
Procedure
1. Ingest & normalize
- Read all
findings/*.md. Parse each finding block (template fields). - Assign/verify stable
FINDING-IDs. Build a table: ID, class, severity, confidence, location,provides,consumes.
2. De-duplicate & cluster
- Merge findings that are the same root cause at the same location reported by two skills (e.g. a reflected value flagged by both XSS and CRLF). Keep one, note both classes. Cluster related findings by shared source or shared sink.
3. Score & apply validation verdicts
- Confirm each finding's standalone severity via
severity-model.md. - Keep
Confidenceorthogonal — never downgrade severity for uncertainty; markTentativeitems for manual confirmation instead. - Read each finding's
Validationverdict (set bysast-finding-validation):- Confirmed (PoC) → keep in the active risk list; these anchor the chains.
- False positive → move to a separate "Validated false positives" section with its FP rationale; exclude from the risk totals and chains.
- Inconclusive / Not validated → keep in the risk list, flagged as "needs manual/dynamic confirmation". Do not silently treat as confirmed.
- If
findings/VALIDATION-LOG.mdexists, cite it. If no validation run happened, say so in §6 and label all findings "static, unvalidated".
4. Build the chain graph (the point of this skill)
These chains are hypotheses (Stage 1) — present them as candidate chains the user can choose to validate (Stage 2), and never imply a graph-derived chain is proven unless a validation run says so.
- For every finding, take its
provides/consumesprimitives. - Draw edge A→B when
provides(A)satisfiesconsumes(B)(per catalog §2/§4). - Match against the named chains (catalog §3) and surface novel ≥2-link paths
ending in a high-impact primitive (
CODE_EXEC,IDENTITY_FORGE, org-wideAUTHZ_BYPASS,SECRET_LEAK→auth). - For each chain: list the ordered links (with finding IDs), the end impact, the single precondition that unlocks it, and the chain severity (uplift = highest impact reachable at the end, even if each link is individually lower).
- Apply chain-validation verdicts (from
sast-finding-validationPhase 3 /findings/VALIDATION-LOG.md). Tag each chain: Validated (end-to-end PoC), Links-confirmed (handoff unproven), Broken (a link is an FP or a handoff fails — drop or downgrade it), or Theoretical (an unvalidated/Inconclusive link). If no chain validation ran, tag every chain Theoretical (static) and say so — never present a graph-derived chain as if it were proven.
5. Prioritize
- Rank by: chain severity → standalone Critical/High → confidence.
- Produce a short "fix these first" list where one fix breaks multiple chains (identify choke-point findings that many chains route through).
Output — SECURITY-ASSESSMENT-REPORT.md
Write to the repo root using this structure:
# Security Assessment Report — <target>
## 1. Executive summary
- Scope, date, stack. Total findings by severity (table). Headline risk in 3 lines:
the worst realistic chain and its business impact.
## 2. Findings summary table
| ID | Class | Severity | Confidence | Validation | Location | Chain(s) |
|----|-------|----------|------------|------------|----------|----------|
## 3. Attack chains ← the centerpiece
For each chain:
### Chain C-<n>: <name> — <Chain severity>
- **Kill chain:** <FINDING-A> (provides X) → <FINDING-B> (consumes X, provides Y) → … ⇒ <end impact>
- **Validation status:** Validated | Links-confirmed | Broken | Theoretical (static)
- **Preconditions:** <the one thing an attacker needs to start>
- **Why it works:** 2-3 sentences tracing the primitive handoffs.
- **Break the chain:** the single highest-leverage fix that severs it.
(Include an ASCII/mermaid-style diagram of the chain graph if more than ~3 chains.)
## 4. Findings by class
One subsection per class that had findings. Full template blocks (or links to the
findings/*.md file) grouped and ordered by severity.
## 5. Choke points & remediation roadmap
- Choke-point findings that appear in multiple chains — fix first.
- Prioritized remediation list (Now / Next / Later) with effort vs risk.
## 6. Validated false positives
Findings that dynamic validation refuted, each with its FP rationale. Excluded
from the risk totals and chains but retained for transparency and audit.
## 7. Methodology & coverage
- Which detection skills ran, whether dynamic validation ran, which classes were
out of scope, and any Tentative/Inconclusive findings needing manual
confirmation. State coverage honestly — do not imply classes were checked if
their skill did not run, or validated if no live run happened.
## 8. Appendix
- Primitive map used, tooling/commands, references (CWE/OWASP), and a pointer to
`findings/VALIDATION-LOG.md` if a validation run occurred.Rules
- Honesty of coverage. If a class skill did not run, say so in §6. Never imply coverage you did not perform.
- Every chain cites real finding IDs. No hypothetical chains without at least one confirmed anchor link; label speculative extensions as such.
- Both severities shown. Per-link severity and chain severity, always.
- English-only. No Turkish anywhere in the report.
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
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.

