vuln-scan

Name: vuln-scan
Author: anthropics

6.7kstars543forks

How to install vuln-scan?

anthropics/defending-code-reference-harness/vuln-scan

$npx -y skills add anthropics/defending-code-reference-harness --skill vuln-scan

Installs into the current project.

Summary

Static source-code vulnerability scan. Reads a target directory (and THREAT_MODEL.md if present), spawns parallel review subagents per focus area, and writes VULN-FINDINGS.json + .md for /triage to consume. Read-only — no building, running, or network. For execution-verified cras

Files · 1

View on GitHub

SKILL.md

1	# /vuln-scan
2
3	Static vulnerability review of a source tree. Produces `VULN-FINDINGS.json`
4	(+ a human-readable `.md`) that `/triage` ingests directly.
5
6	This skill does not execute code. It reads source and reasons about it.
7	For execution-verified findings (ASAN crashes, reproducing PoCs), point the
8	user at `vuln-pipeline run <target>` — see README Step 2.
9
10	Tool fallbacks. Prefer the dedicated Glob and Grep tools. Some sessions
11	do not provision them — `allowed-tools` is a permission filter, not a loader,
12	so listing them here does not make them appear. When Glob/Grep are
13	unavailable, fall back to the read-only Bash commands whitelisted above:
14	`rg --files <scope>` / `ls -R` for enumeration, `rg -n` / `grep -rn` for
15	search, `wc` / `head` / `file` for sniffing. These are the ONLY permitted
16	Bash commands; do not write helper scripts or pipe target content into a
17	shell interpreter.
18
19	## Arguments
20
21	- `<target-dir>` (required) — directory to scan. Relative or absolute.
22	- `--focus <area>` — scan only this focus area (repeatable). Skips recon.
23	- `--single` — no subagent fan-out; one sequential pass. Use on tiny targets
24	or when debugging the prompt.
25	- `--extra <file>` — append the contents of `<file>` to the review brief
26	(after the category list). Use to add org-specific vulnerability classes,
27	compliance checks, or stack-specific patterns. Plain text; same shape as
28	the category blocks below.
29	- `--no-score` — skip the Step 3b confidence pass (saves a round of
30	subagents). Findings keep the scanner's self-reported confidence only.
31
32	## Step 1 — Scope
33
34	1. Resolve `<target-dir>`. If it doesn't exist or has no source files, stop
35	with an error.
36	2. Look for `<target-dir>/THREAT_MODEL.md`. If present, parse its section 3 "Entry
37	points & trust boundaries" table and section 4 "Threats" table for focus areas
38	and threat classes. This is the preferred scoping input.
39	3. If no THREAT_MODEL.md and no `--focus`: do a quick recon — list the
40	source tree, read entry points and dispatch code, and propose 3-10 focus
41	areas using the pattern `<subsystem> (<function/file>) — <key operations>`.
42	Same shape as `harness/prompts/recon_prompt.py`.
43	4. If `--focus` was given, use exactly those.
44
45	Tell the user the focus areas you'll scan and the source-file count before
46	fanning out.
47
48	## Step 2 — Fan out
49
50	Unless `--single`, spawn one Task subagent per focus area in parallel.
51	Cap at 10 concurrent. Each subagent gets the review brief below with its
52	focus area filled in. On tiny targets (<15 source files), fall through to
53	`--single` automatically.
54
55	### Review brief (per subagent)
56
57	```
58	You are conducting authorized static security review of source code. Your
59	focus area: {focus_area}. Other agents cover other areas; duplication
60	is wasted effort.
61
62	TARGET: {target_dir}
63	TRUST BOUNDARY: {from THREAT_MODEL.md section 3, or "untrusted input → process memory"}
64
65	TASK: read the source in your focus area and identify candidate
66	vulnerabilities. This is static review — do NOT build, run, or probe
67	anything. Reason from the code.
68
69	REPORTING BAR: report anything with a plausible exploit path. Skip style
70	concerns, best-practice gaps, and purely theoretical issues with no attack
71	story at all — but if you're unsure whether something is real, REPORT IT
72	with a low confidence score rather than dropping it. A downstream triage
73	step does the rigorous verification; your job is to not miss things.
74
75	WHAT TO LOOK FOR:
76
77	MEMORY SAFETY (C/C++ and unsafe/FFI blocks) — HIGH VALUE:
78	- heap-buffer-overflow / stack-buffer-overflow / global-buffer-overflow
79	- heap-use-after-free / double-free
80	- integer overflow feeding an allocation or index
81	- format-string bugs
82	- unbounded recursion or allocation driven by untrusted size fields
83
84	INJECTION & CODE EXECUTION — HIGH VALUE:
85	- SQL / command / LDAP / XPath / NoSQL / template injection
86	- path traversal in file operations
87	- unsafe deserialization (pickle, YAML, native), eval injection
88	- XSS (reflected, stored, DOM-based) — but see React/Angular note below
89
90	AUTH, CRYPTO, DATA — HIGH VALUE:
91	- authentication or authorization bypass, privilege escalation
92	- TOCTOU on a security check
93	- hardcoded secrets, weak crypto, broken cert validation

People who liked this also viewed

6 picks

Type

vuln-scan

6.7kstars543forks

View on GitHub

How to install vuln-scan?

anthropics/defending-code-reference-harness/vuln-scan

$npx -y skills add anthropics/defending-code-reference-harness --skill vuln-scan

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# /vuln-scan
2
3	Static vulnerability review of a source tree. Produces `VULN-FINDINGS.json`
4	(+ a human-readable `.md`) that `/triage` ingests directly.
5
6	This skill does not execute code. It reads source and reasons about it.
7	For execution-verified findings (ASAN crashes, reproducing PoCs), point the
8	user at `vuln-pipeline run <target>` — see README Step 2.
9
10	Tool fallbacks. Prefer the dedicated Glob and Grep tools. Some sessions
11	do not provision them — `allowed-tools` is a permission filter, not a loader,
12	so listing them here does not make them appear. When Glob/Grep are
13	unavailable, fall back to the read-only Bash commands whitelisted above:
14	`rg --files <scope>` / `ls -R` for enumeration, `rg -n` / `grep -rn` for
15	search, `wc` / `head` / `file` for sniffing. These are the ONLY permitted
16	Bash commands; do not write helper scripts or pipe target content into a
17	shell interpreter.
18
19	## Arguments
20
21	- `<target-dir>` (required) — directory to scan. Relative or absolute.
22	- `--focus <area>` — scan only this focus area (repeatable). Skips recon.
23	- `--single` — no subagent fan-out; one sequential pass. Use on tiny targets
24	or when debugging the prompt.
25	- `--extra <file>` — append the contents of `<file>` to the review brief
26	(after the category list). Use to add org-specific vulnerability classes,
27	compliance checks, or stack-specific patterns. Plain text; same shape as
28	the category blocks below.
29	- `--no-score` — skip the Step 3b confidence pass (saves a round of
30	subagents). Findings keep the scanner's self-reported confidence only.
31
32	## Step 1 — Scope
33
34	1. Resolve `<target-dir>`. If it doesn't exist or has no source files, stop
35	with an error.
36	2. Look for `<target-dir>/THREAT_MODEL.md`. If present, parse its section 3 "Entry
37	points & trust boundaries" table and section 4 "Threats" table for focus areas
38	and threat classes. This is the preferred scoping input.
39	3. If no THREAT_MODEL.md and no `--focus`: do a quick recon — list the
40	source tree, read entry points and dispatch code, and propose 3-10 focus
41	areas using the pattern `<subsystem> (<function/file>) — <key operations>`.
42	Same shape as `harness/prompts/recon_prompt.py`.
43	4. If `--focus` was given, use exactly those.
44
45	Tell the user the focus areas you'll scan and the source-file count before
46	fanning out.
47
48	## Step 2 — Fan out
49
50	Unless `--single`, spawn one Task subagent per focus area in parallel.
51	Cap at 10 concurrent. Each subagent gets the review brief below with its
52	focus area filled in. On tiny targets (<15 source files), fall through to
53	`--single` automatically.
54
55	### Review brief (per subagent)
56
57	```
58	You are conducting authorized static security review of source code. Your
59	focus area: {focus_area}. Other agents cover other areas; duplication
60	is wasted effort.
61
62	TARGET: {target_dir}
63	TRUST BOUNDARY: {from THREAT_MODEL.md section 3, or "untrusted input → process memory"}
64
65	TASK: read the source in your focus area and identify candidate
66	vulnerabilities. This is static review — do NOT build, run, or probe
67	anything. Reason from the code.
68
69	REPORTING BAR: report anything with a plausible exploit path. Skip style
70	concerns, best-practice gaps, and purely theoretical issues with no attack
71	story at all — but if you're unsure whether something is real, REPORT IT
72	with a low confidence score rather than dropping it. A downstream triage
73	step does the rigorous verification; your job is to not miss things.
74
75	WHAT TO LOOK FOR:
76
77	MEMORY SAFETY (C/C++ and unsafe/FFI blocks) — HIGH VALUE:
78	- heap-buffer-overflow / stack-buffer-overflow / global-buffer-overflow
79	- heap-use-after-free / double-free
80	- integer overflow feeding an allocation or index
81	- format-string bugs
82	- unbounded recursion or allocation driven by untrusted size fields
83
84	INJECTION & CODE EXECUTION — HIGH VALUE:
85	- SQL / command / LDAP / XPath / NoSQL / template injection
86	- path traversal in file operations
87	- unsafe deserialization (pickle, YAML, native), eval injection
88	- XSS (reflected, stored, DOM-based) — but see React/Angular note below
89
90	AUTH, CRYPTO, DATA — HIGH VALUE:
91	- authentication or authorization bypass, privilege escalation
92	- TOCTOU on a security check
93	- hardcoded secrets, weak crypto, broken cert validation

People who liked this also viewed

6 picks

Type