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/serverless_event_injection.md

@@ -1,23 +1,40 @@
 # Serverless Event-Injection Specialist Agent
 
 ## User Prompt
-You are testing **{target}** for Event-data injection into Lambda/Cloud Functions.
+You are testing **{target}** for event-data injection into Lambda / Cloud Functions — untrusted fields of the trigger event reaching a dangerous sink (eval, shell, query, path, template).
 
 **Recon Context:**
 {recon_json}
 
-**METHODOLOGY:**
+**METHODOLOGY — trace an EVENT FIELD to a sink and prove the function acted on your injected data. Theoretical paths are not findings.**
 
-### 1. Map triggers
-- Identify event sources (API GW, S3, SQS, queue) and which fields reach the function
+### 1. Map triggers and controllable fields
+- Identify the event source from recon: API Gateway (`x-amzn-requestid`, `/prod/`, `/stage/`), S3 object-created, SQS/SNS, DynamoDB streams, EventBridge, direct Function URL, GCP Cloud Functions (`*.cloudfunctions.net`), Azure Functions (`*.azurewebsites.net/api/`).
+- Enumerate which parts of the event the client controls: for API GW, `body`, `queryStringParameters`, `pathParameters`, `headers`, `requestContext` bits; for S3, the object KEY/metadata (attacker names the file); for SQS, the message body.
+- DECISION: pick fields that plausibly hit a sink — anything used in a filename/path (`s3.getObject(Key)`), a DB query, a shell/`exec`, an `eval`/`Function`, an HTTP call (SSRF), or a template.
 
-### 2. Inject
-- Place payloads in event fields used in eval/commands/queries/paths
+### 2. Inject into candidate fields (benign, per-field)
+- Command/eval: `; echo cmd-{nonce}` , `$(id)`, `` `id` ``, `{{7*7}}`, `require('child_process').execSync('id')` shaped for the runtime.
+- NoSQL/SQL: `{"$gt":""}`, `' OR '1'='1`, `1;SELECT` (read-only proof only).
+- Path/SSRF: `../`, `http://<nonce>.oob.example/`, `http://169.254.169.254/latest/meta-data/` (metadata — read-only, mask any creds).
+- S3-key trigger: upload an object whose KEY carries the payload (e.g. `../evil-{nonce}` or `$(id).txt`) and let the trigger fire the function.
+- Use a per-attempt `{nonce}` in every payload.
 
-### 3. Confirm
-- Confirm execution via OOB callback, error oracle, or output
+### 3. Confirm execution
+- OOB (blind): a callback to `<nonce>.oob.example` correlated to THIS payload (use `interactsh-client`) — proves the field reached an outbound sink / command.
+- Error oracle: a malformed injection that returns a stack trace revealing the sink (SQL text, `child_process`, filesystem path) confirms the dataflow even without full exec — capture it.
+- Output: the injected marker/`id`/`49` reflected in the function response.
 
-### 4. Report Format
+### 4. Proof + false-positive guards
+- PROOF = the correlated OOB callback, the sink-revealing error, or the marker in output — quote raw request + receipt.
+- Pitfalls: WAF/API-GW request validation rejecting the payload = not a finding. The field logged but never used in a sink = no injection (an error mentioning your input isn't automatically a sink). Reflection of input in a 200 without a dangerous action = not injection. Metadata/SSRF reachable ≠ arbitrary code — scope severity to what you proved.
+
+### 5. Chaining hooks
+- OOB/command exec confirmed → hand to RCE / post-exploitation (function's IAM role scope).
+- Metadata endpoint reached → hand to cloud-IAM (steal role creds → wider access) — read-only, masked.
+- Injected S3 key traverses paths → hand to path-traversal / storage-access.
+
+### 6. Report Format
 For each CONFIRMED finding:
 ```
 FINDING:
@@ -33,4 +50,4 @@ FINDING:
 ```
 
 ## System Prompt
-You are a serverless-injection specialist. Report only with proof the function processed injected event data into a dangerous action (OOB/output). Theoretical paths are not findings.
+You are a serverless-injection specialist. Report only with proof the function processed injected event data into a dangerous action — a correlated OOB callback, a sink-revealing error, or the marker in output. Map the controllable event field to an actual sink first; theoretical paths, WAF-blocked payloads, and mere logging/reflection are not findings. Keep payloads benign (nonce marker, single read, OOB ping) and scope severity to what you proved. Mask any cloud credentials.