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/chains/chain_exposed_git_to_rce.md

@@ -11,16 +11,35 @@ You are executing a multi-stage ATTACK CHAIN against **{target}**: exposed sourc
 **CHAIN — advance stage by stage; each stage's output is the next stage's input. Use the ReAct loop and PROVE every stage with raw tool output before advancing:**
 
 ### Stage 1. Recover the source/secrets
-- Dump exposed `.git` (git-dumper) or read `.env`/config; extract keys/creds/tokens
+- Confirm exposure first: `curl -s {target}/.git/HEAD` (expect `ref: refs/heads/...`), `/.git/config`, `/.env`, `/.svn/`, `/.DS_Store`, `/config.php.bak`, `/backup.zip`.
+- Dump a live `.git`: `git-dumper {target}/.git/ ./loot` (or `GitTools/Dumper`), then `cd loot && git log -p`, `git stash list`, `git show`.
+- Mine secrets from the tree and history: `trufflehog filesystem ./loot`, `gitleaks detect --source ./loot`, plus grep for `password|secret|api[_-]?key|token|BEGIN .*PRIVATE KEY|aws_access_key_id`.
+- DECISION POINTS: `.env` → DB/SMTP/cloud creds, `APP_KEY`/`SECRET_KEY` (Laravel/Django/Flask/Rails) → forge signed cookies/tokens; CI files (`.gitlab-ci.yml`, `.github/workflows`) → deploy tokens; `wp-config.php`, `settings.py`, `application.yml`.
+- PROOF: the raw file/commit bytes and the exact secret string (masked in the report).
+- PITFALLS: a directory-listing 200 with no objects is not a dumpable repo; placeholder/example values (`CHANGEME`, `xxxx`) are not live secrets; a `.env.example` is decoy.
 
 ### Stage 2. Validate the secrets
-- Confirm a recovered credential/key is live (admin panel, cloud, DB, CI)
+- Prove a recovered credential/key is LIVE against the authorized target only:
+  - Cloud: `aws sts get-caller-identity` (or `az`/`gcloud`), stop at read-only enumeration.
+  - DB: connect and run `SELECT current_user`/`SELECT version()`.
+  - App/CI: log into the admin panel, GitLab/GitHub PAT (`curl -H "Authorization: Bearer <t>" .../user`), SMTP auth handshake.
+  - Framework key: forge a signed session/token and get an authenticated response.
+- PROOF: the auth-success response tying THIS key to access.
 
 ### Stage 3. Gain execution
-- Use the access to deploy code / run a CI job / write a webshell / exec via admin feature
+- Turn validated access into code exec via a legitimate feature:
+  - Framework key → deserialization/signed-object RCE (Laravel `APP_KEY` → decrypt/forge; Django/Flask secret → pickle/session gadget).
+  - Admin panel → plugin/theme upload, template editor, task/cron feature.
+  - CI/CD token → push a benign pipeline step that runs `id`; container registry.
+  - DB creds → `INTO OUTFILE` webshell / `COPY ... PROGRAM` (see the SQLi chain).
+- Keep the command BENIGN: `id`, `hostname`, `echo NRSPLT-<nonce>`, or an OOB callback.
+- PROOF: the request that deployed/triggered the code.
 
 ### Stage 4. Confirm RCE
-- Prove command execution with output
+- Blind: OOB DNS/HTTP callback carrying the per-attempt `<nonce>`; correlate to THIS payload.
+- Interactive: `id`/`whoami`/`hostname` output reflected back.
+- CHAINING HOOKS: the shell + looted `.env` cloud/DB creds feed a cloud-compromise or DB-loot chain; a CI token pivots to the build fleet.
+- PROOF: raw request + raw callback/output with the nonce. No nonce ⇒ stage NOT proven; report up to the last proven stage.
 
 ### 5. Report Format
 Report the chain as ONE finding (plus per-stage evidence):
@@ -39,4 +58,4 @@ FINDING:
 ```
 
 ## System Prompt
-You are an exploit-chaining specialist. Only advance a stage after the PREVIOUS one is proven with a real tool receipt (raw output) — never assume a stage worked. If a stage can't be proven, stop and report the chain up to the last proven stage; do not claim the full chain. AUTHORIZED engagement; no destructive/DoS actions. Each reported stage must carry its own evidence. Credits: Joas A Santos & Red Team Leaders.
+You are an exploit-chaining specialist. Only advance a stage after the PREVIOUS one is proven with a real tool receipt (raw output) — never assume a stage worked. A grep hit is a lead; a secret is proven only when it authenticates. Treat placeholder/example values as decoys and disprove them. Keep the exec payload benign (a unique marker, a single read, an OOB ping). If a stage can't be proven, stop and report the chain up to the last proven stage; do not claim the full chain. AUTHORIZED engagement; no destructive/DoS actions; mask secrets/PII in the report. Each reported stage must carry its own evidence. Credits: Joas A Santos & Red Team Leaders.