Authenticated Session Acquisition

On financial and multi-tenant targets the login is gated by SMS OTP / TOTP MFA that an autonomous run cannot satisfy, so the post-auth surface (BOLA/IDOR/mass-assignment/injection on the real data APIs, session handling, API pivots) silently collapses to pre-auth findings only. This skill's job is narrow and concrete: get one legitimate authenticated session and hand it to the executors as a reusable artifact. It does not attack the auth mechanism — see authentication for that.

The artifact contract (how the session reaches executors)

Emit both, into the engagement's OUTPUT_DIR, referenced BY NAME (never inline secrets — credential-loading.md):

• OUTPUT_DIR/<asset>/session/storageState.json — the Playwright storage state (cookies + localStorage) for browser-driven post-auth testing.
• OUTPUT_DIR/<asset>/session/bearer.txt — the raw access/ID token (+ its expires_at) for direct API replay.

Executors consume these exactly like any env-loaded secret: the scope file's creds_env names the realm, and the session path is passed as a file reference, so no token is ever written into a prompt, experiments.md, or attack-chain.md. Re-run this skill when the token expires (record expires_at; refresh rather than re-login where a refresh token exists).

Modes (pick by what the client can provide)

Provider recipes

Detailed per-IdP flows are in reference/session-acquisition.md: Auth0, Okta, Amazon Cognito (SRP auth + the UNSIGNED unauthenticated-posture check), and Descope. Each recipe ends by exporting storageState.json + bearer.txt in the contract above.

When a session cannot be acquired — defer honestly, do not fake

If no test account / OTP seed is available and no OTP can be relayed, the realm's post-auth surface is untestable through no fault of the tester. Record it honestly instead of a fabricated NA or a silent miss:

1. File a client-input request at reports/client-input-requests/CIR-NNN.md naming exactly what's needed ("needs test account / OTP seed / allowlist for <realm>").
2. Set a realm-level BLOCKED_REASON in attack-chain.md.
3. Mark that realm's post-auth coverage cells status:"deferred" in coverage.json, each carrying deferral_reason + client_input_request (the CIR path) — see coverage-matrix.md. The deterministic gate then discloses the deferred surface in the report rather than hiding it, and only the parent orchestrator (via coverage_gate.py --accept-deferrals) may finalize a substantiated-deferred engagement. A scope that is entirely auth-gated never completes.

The preflight cred-reach probe (preflight-checklist.md, Phase-1 gate) is what triggers this path: present creds are not working creds; verify reachability per realm before spawning post-auth executors.

Workflow

1. Read the scope: which realms/tenants are in creds_env, which assets are behind them, roe.reversible_writes for the test tenant.
2. Identify the IdP (login redirect host / /.well-known/openid-configuration / SDK bundle) → pick the recipe.
3. Acquire the session in the appropriate mode → write storageState.json + bearer.txt.
4. Verify: one minimal authenticated request returns an authorized response (not 401/403/login-redirect).
5. Hand the artifact paths to the post-auth executor batch; on failure, run the defer path above.

Anti-Patterns

• Do not paste tokens, seeds, or cookies into prompts, experiments.md, attack-chain.md, or any tools/*.md — reference the artifact file by path only.
• Do not brute-force or bypass the OTP/MFA challenge here — that is an authentication finding, not session acquisition.
• Do not mark an MFA-blocked cell covered/NA — mark it deferred with a filed CIR so the gate discloses it.
• Do not test outside the authenticated tenant you were given, and honor roe.reversible_writes (create-then-delete only in your own test tenant).

Reference

• reference/session-acquisition.md — per-IdP acquisition recipes (Auth0 / Okta / Cognito / Descope) + the storageState/Bearer export.
• tools/totp_now.py — stdlib RFC-6238 TOTP generator (the TOTP-from-seed mode).