finding-discipline

Name: finding-discipline
Author: hypnguyen1209

320stars55forks

How to install finding-discipline?

hypnguyen1209/offensive-claude/finding-discipline

$npx -y skills add hypnguyen1209/offensive-claude --skill finding-discipline

Installs into the current project.

Summary

Use when about to record, claim, rate the severity of, or report any security finding — before marking anything [CONFIRMED] or writing it into the report

Files · 1

View on GitHub

SKILL.md

1	# Finding Discipline
2
3	## Overview
4
5	The Iron Law: No `[CONFIRMED]` without proof. A finding is a claim about real, demonstrated
6	impact — not a status code, a reflected string, or a hunch. This is the offensive equivalent of
7	test-driven development: the proof is the test, and the finding does not exist until it passes.
8
9	Violating the letter of this rule is violating its spirit.
10
11	## The three tiers
12
13	- `[CONFIRMED]` — impact demonstrated AND grounded in an evidence artifact that exists on disk.
14	- `[POSSIBLE]` — reachable but the class proof bar is not yet met. Keep digging; do not report as real.
15	- `[INFO]` — true but no security impact at the claimed severity.
16
17	## The proof bar (per class)
18
19	A status code is not impact. See `skills/references/finding-evidence-standards.md`:
20	SSRF needs the internal response; IDOR needs another principal's data; RCE needs command output;
21	XSS needs script execution; open redirect needs an external destination.
22
23	## Required mechanical gate
24
25	Before any finding is recorded or reported:
26
27	1. Set the structured proof signal and run
28	`python skills/vulnerability-analysis/scripts/validate_findings.py --findings f.json --evidence ./evidence`
29	— it rejects ungrounded findings and per-class false positives.
30	2. Pass it through the REQUIRED `finding-validator` agent (PASS / KILL / DOWNGRADE).
31	3. Only a `[CONFIRMED]` survivor goes in the report. Severity must match what was demonstrated (CVSS honest).
32
33	See `skills/references/finding-validation-runtime.md` for the 7-question gate.
34
35	## Read-first, never name-guess
36
37	If a function calls another, read the callee — do not infer what it does from its name.
38	`sanitize()`, `is_safe()`, `validate_input()` are exactly where vulnerabilities hide; a name is a
39	claim by the author, not evidence. The same applies to a config key, a decorator, or a CVE title:
40	open the thing before you reason about it. An unread dependency in a data-flow claim is a hole.
41
42	## Quote-grounded confidence tiers
43
44	Every claim in a finding carries a confidence grounded in what you can quote, not in how it feels:
45
46	- High — a direct quote from the artifact (the exact source line, the response body, the
47	decompiled instruction) supports the claim.
48	- Medium — an explicitly stated assumption bridges a gap you could not directly observe.
49	State the assumption; do not hide it.
50	- Low — an inference you are flagging as unverified. Acceptable to record, never to report as
51	fact.
52
53	This is orthogonal to the `[CONFIRMED]/[POSSIBLE]/[INFO]` tier (which is about impact); confidence
54	is about how well-grounded the claim is. A High-confidence claim of a Low-impact bug is still Low.
55
56	## Red Flags — STOP, downgrade to `[POSSIBLE]`
57
58	- "I'm sure it's exploitable" (but haven't shown impact)
59	- "The payload reflected, so it's XSS" / "I got a 200, so it works"
60	- "The DNS callback fired" (SSRF with no internal response)
61	- "I changed the id and got data" (but it was my own — self-IDOR)
62	- "It probably works on a real target" (default-deployment unproven)
63
64	All of these mean: it is `[POSSIBLE]`, not `[CONFIRMED]`. Get the evidence.
65
66	## Rationalizations
67
68	\| Excuse \| Reality \|
69	\|--------\|---------\|
70	\| "Pattern-matched, clearly vulnerable" \| Pattern ≠ exploit. Demonstrate impact or it's POSSIBLE. \|
71	\| "I'll attach evidence later" \| No artifact on disk = ungrounded = REJECTED now. \|
72	\| "Blind RCE counts" \| No output / no OOB confirmation = POSSIBLE, not CONFIRMED. \|
73	\| "Severity is obviously Critical" \| CVSS must reflect demonstrated impact, not the worst case. \|
74	\| "The validator is overkill here" \| Untested findings are how false positives ship. Run it. \|
75	\| "The function is called `sanitize`, so it's safe" \| A name is the author's claim, not proof. Read the callee. \|
76	\| "The solver timed out, so it's not exploitable" \| A tool limit is `feasibility:null` (manual), never `false`. \|
77
78	A short list of CONFIRMED findings beats a long list of POSSIBLEs. Killing a false positive is success.

People who liked this also viewed

6 picks

Type

entra-app-registrationGuides Microsoft Entra ID app registration, OAuth 2.0 authentication, and MSAL integration. USE FOR: create app registration, register Azure AD app, configure OAuth, set up authentication, add API perSkillsJul 2026471k1.3k
azure-complianceRun Azure compliance and security audits with azqr plus Key Vault expiration checks. Covers best-practice assessment, resource review, policy/compliance validation, and security posture checks. WHEN: SkillsJul 2026470k1.3k
azure-rbacOfficial agent plugin providing skills and MCP server configurations for Azure scenarios.SkillsJul 2026466k1.3k
better-auth-best-practicesAgent skill from better-auth/skillsSkillsJul 202678k203
firebase-security-rules-auditorA skill to evaluate how secure Firestore security rules are. Use this when Firestore security rules are updated to ensure that the generated rules are extremely secure and robust.SkillsJul 202676k386
audit-websiteAudit a website with the squirrelscan CLI and fix the findings in code. Runs SEO, performance, security, technical, content, accessibility, and 15 other rule categories (260+ rules), returns an LLM-opSkillsJul 202665k85

finding-discipline

320stars55forks

View on GitHub

How to install finding-discipline?

hypnguyen1209/offensive-claude/finding-discipline

$npx -y skills add hypnguyen1209/offensive-claude --skill finding-discipline

Installs into the current project.

Summary

Use when about to record, claim, rate the severity of, or report any security finding — before marking anything [CONFIRMED] or writing it into the report

Files · 1

View on GitHub

SKILL.md

1	# Finding Discipline
2
3	## Overview
4
5	The Iron Law: No `[CONFIRMED]` without proof. A finding is a claim about real, demonstrated
6	impact — not a status code, a reflected string, or a hunch. This is the offensive equivalent of
7	test-driven development: the proof is the test, and the finding does not exist until it passes.
8
9	Violating the letter of this rule is violating its spirit.
10
11	## The three tiers
12
13	- `[CONFIRMED]` — impact demonstrated AND grounded in an evidence artifact that exists on disk.
14	- `[POSSIBLE]` — reachable but the class proof bar is not yet met. Keep digging; do not report as real.
15	- `[INFO]` — true but no security impact at the claimed severity.
16
17	## The proof bar (per class)
18
19	A status code is not impact. See `skills/references/finding-evidence-standards.md`:
20	SSRF needs the internal response; IDOR needs another principal's data; RCE needs command output;
21	XSS needs script execution; open redirect needs an external destination.
22
23	## Required mechanical gate
24
25	Before any finding is recorded or reported:
26
27	1. Set the structured proof signal and run
28	`python skills/vulnerability-analysis/scripts/validate_findings.py --findings f.json --evidence ./evidence`
29	— it rejects ungrounded findings and per-class false positives.
30	2. Pass it through the REQUIRED `finding-validator` agent (PASS / KILL / DOWNGRADE).
31	3. Only a `[CONFIRMED]` survivor goes in the report. Severity must match what was demonstrated (CVSS honest).
32
33	See `skills/references/finding-validation-runtime.md` for the 7-question gate.
34
35	## Read-first, never name-guess
36
37	If a function calls another, read the callee — do not infer what it does from its name.
38	`sanitize()`, `is_safe()`, `validate_input()` are exactly where vulnerabilities hide; a name is a
39	claim by the author, not evidence. The same applies to a config key, a decorator, or a CVE title:
40	open the thing before you reason about it. An unread dependency in a data-flow claim is a hole.
41
42	## Quote-grounded confidence tiers
43
44	Every claim in a finding carries a confidence grounded in what you can quote, not in how it feels:
45
46	- High — a direct quote from the artifact (the exact source line, the response body, the
47	decompiled instruction) supports the claim.
48	- Medium — an explicitly stated assumption bridges a gap you could not directly observe.
49	State the assumption; do not hide it.
50	- Low — an inference you are flagging as unverified. Acceptable to record, never to report as
51	fact.
52
53	This is orthogonal to the `[CONFIRMED]/[POSSIBLE]/[INFO]` tier (which is about impact); confidence
54	is about how well-grounded the claim is. A High-confidence claim of a Low-impact bug is still Low.
55
56	## Red Flags — STOP, downgrade to `[POSSIBLE]`
57
58	- "I'm sure it's exploitable" (but haven't shown impact)
59	- "The payload reflected, so it's XSS" / "I got a 200, so it works"
60	- "The DNS callback fired" (SSRF with no internal response)
61	- "I changed the id and got data" (but it was my own — self-IDOR)
62	- "It probably works on a real target" (default-deployment unproven)
63
64	All of these mean: it is `[POSSIBLE]`, not `[CONFIRMED]`. Get the evidence.
65
66	## Rationalizations
67
68	\| Excuse \| Reality \|
69	\|--------\|---------\|
70	\| "Pattern-matched, clearly vulnerable" \| Pattern ≠ exploit. Demonstrate impact or it's POSSIBLE. \|
71	\| "I'll attach evidence later" \| No artifact on disk = ungrounded = REJECTED now. \|
72	\| "Blind RCE counts" \| No output / no OOB confirmation = POSSIBLE, not CONFIRMED. \|
73	\| "Severity is obviously Critical" \| CVSS must reflect demonstrated impact, not the worst case. \|
74	\| "The validator is overkill here" \| Untested findings are how false positives ship. Run it. \|
75	\| "The function is called `sanitize`, so it's safe" \| A name is the author's claim, not proof. Read the callee. \|
76	\| "The solver timed out, so it's not exploitable" \| A tool limit is `feasibility:null` (manual), never `false`. \|
77
78	A short list of CONFIRMED findings beats a long list of POSSIBLEs. Killing a false positive is success.

People who liked this also viewed

6 picks

Type