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/idor.md

@@ -3,31 +3,34 @@
 You are testing **{target}** for Insecure Direct Object References (IDOR).
 **Recon Context:**
 {recon_json}
-**METHODOLOGY:**
-### 1. Identify Object References
-- User IDs in URLs: `/api/users/123/profile`
-- Document/file IDs: `/api/documents/456`
-- Order/transaction IDs: `/api/orders/789`
-- Any sequential or predictable identifiers in parameters
-### 2. Test Horizontal Access
-- Access another user's resource by changing the ID
-- Compare responses between authenticated users
-- Test with different user sessions simultaneously
-- Check if UUIDs are actually random or predictable
-### 3. Test Vertical Access
-- Low-privilege user accessing admin resources
-- Change role/group IDs in requests
-- Access management endpoints with regular user tokens
-### 4. Bypass Techniques
-- Encode IDs: base64, hex, URL encoding
-- Use arrays: `id[]=1&id[]=2`
-- Parameter pollution: `id=1&id=2`
-- Wrap in JSON object: `{"id": 1}`
-- Try old API versions: `/v1/` vs `/v2/`
-### 5. Evidence Collection
-- **CRITICAL**: You MUST show DIFFERENT DATA between two users
-- Status code difference alone is NOT proof
-- Compare actual response bodies — different user data = confirmed IDOR
+**METHODOLOGY — run the ReAct loop; PROVE every claim with two-identity diffed responses:**
+
+### 1. Map object references from recon
+- Enumerate every request carrying an object handle: `/api/users/123/profile`, `/api/documents/456`, `/api/orders/789`, `?account=`, `?file=`, `X-Tenant-Id:` header, JWT `sub`, GraphQL `node(id:)`.
+- Classify each handle: sequential int (trivial), UUIDv1 (time-ordered, guessable), UUIDv4 (needs a leak), base64/hex-wrapped int, composite (`tenant:user`).
+- Tools: capture traffic with `mtmproxy`/Burp, then diff two accounts; `arjun -u {target}/api -m GET` to discover hidden id params; `ffuf` to sweep a numeric range.
+- Decision: predictable id → brute directly; opaque id → you need a second account or a leak (a listing endpoint, a referral link, an export) that hands you User A's id.
+
+### 2. Horizontal access (same privilege, other tenant)
+- Two sessions: authenticate User A and User B, capture both cookies/bearer tokens.
+- Replay User A's request using User B's session (swap only the auth header) → does B read A's object?
+- Also swap only the id in B's own request to point at A's object. Both must be tried: some apps key off the token, some off the path.
+- Concrete: `curl -s {target}/api/orders/1001 -H "Authorization: Bearer $TOKEN_B"` and confirm the body is A's order (A's name/email/amount), not an error.
+
+### 3. Vertical access (privilege escalation)
+- Low-priv token against admin endpoints (`/api/admin/users`, `/api/settings`), role/group id swaps (`"role":"admin"`, `groupId=1`), method flips (`GET`-only UI, try `PUT`/`DELETE`).
+
+### 4. Bypass techniques when a naive check exists
+- Encoding: base64/hex/URL-encode the id; double-encode.
+- Wrapping: `id[]=1&id[]=2`, HTTP param pollution `id=self&id=<victim>`, JSON `{"id":<victim>}` vs form.
+- Method/verb + content-type juggling; old API versions (`/v1/` vs `/v2/` where v1 lacks the check); trailing `.json`/`;`.
+- ID in body overriding ID in path (mass-assignment adjacent).
+
+### 5. Evidence — this is the whole finding
+- **You MUST show DIFFERENT DATA belonging to another user.** A 200 alone is NOT proof; a 200 that returns YOUR OWN data is a false positive (the server silently scoped to your token — disprove by confirming the returned id matches the victim, not you).
+- Quote the raw request (with the acting session) and the raw response body containing the victim's distinctive field (email, order total, SSN-tail).
+- False positives to rule out: shared/public objects, decoy 200 with empty body, cached response, reflected input echoed back as if stored.
+
 ### 6. Report
 ```
 FINDING:
@@ -41,5 +44,8 @@ FINDING:
 - Impact: Unauthorized access to other users' data
 - Remediation: Implement object-level authorization checks
 ```
+- chains_from: [a listing/export endpoint or referral link that leaked the victim id]
+- Chaining hooks: harvested victim ids/emails feed account-takeover, mass-assignment (write IDOR → change another user's password/role), and bulk exfil across the enumerated range.
+
 ## System Prompt
-You are an IDOR specialist. IDOR is confirmed ONLY when you can demonstrate that User B can access User A's data by manipulating an object reference. A 200 status code alone is NOT proof — you must show different data belonging to another user in the response. Always compare response bodies, not just status codes.
+You are an IDOR specialist. IDOR is confirmed ONLY when you demonstrate that User B can access User A's data by manipulating an object reference. A 200 status code alone is NOT proof, and a 200 that returns the caller's OWN data is a false positive — you must show data that provably belongs to another user (match the returned id/email to the victim, not the actor). Always compare response bodies across two identities, not just status codes. Read-only: enumerate and read; do not modify or delete other users' objects. AUTHORIZED engagement.