Program version: 1.0 Effective: 2026-11-14 Policy owner: security@seele.agency Canonical URL: https://seele.agency/security Machine-readable: https://seele.agency/.well-known/security.txt
Read this first. The program described below is fully specified but NOT yet funded or running. As of 2026-11-14:
- The revenue source (
xkg-payments) has no real Stripe keys — only placeholders (sk_test_replace,sk_live_replace) — and the service is not running.- Consequently there is no revenue, therefore no 3% allocation, and therefore no pool balance. The pool balance is $0.
- No bounty payout can be made today. No researcher can receive money under this program until the activation checklist in §9 is complete.
What exists today: the policy, the payout tiers, the response SLAs, the safe-harbor language, the public disclosure pages, and the ledger format. Reports are still accepted and still triaged and fixed — the bounty is the only part that is pending.
This distinction matters for legal clarity: accepting reports and fixing vulnerabilities is a security practice. A bounty is a financial commitment. We are not yet making the financial commitment, and we do not want researchers to assume otherwise.
Seele Agency will pay security researchers who report qualifying vulnerabilities in Seele Agency software. The program is funded from a pool that accrues 3% of net revenue (see §4).
Findings are handled under a coordinated disclosure model. We ask for 90 days before public disclosure, and we commit to acknowledging reports within 3 business days.
Every report is treated as a security issue first and a bounty candidate second. A report that is valid and useful will be:
If the pool is empty, the report is still accepted, still triaged, and still fixed, and the researcher is credited on /security/acknowledgments/ if they consent. No researcher is ever told "no bounty" before being told "this is a real bug."
| Asset | Hostname / location | Notes |
|---|---|---|
| Marketing + docs site | https://seele.agency, https://www.seele.agency |
Static site deployed via Cloudflare Pages (project seele-agency, prod branch master) |
| Public docs site | https://docs.seele.agency |
Docusaurus 3.10.2 |
| Worker (public) | https://seele-memory.griptoad.workers.dev |
Public workers.dev hostname — see §2.3. No Cloudflare Access bound. |
| Worker source | seele-agency-site/cf-workers/ |
Proxy worker; ORIGIN_URL configurable |
| Payments service | xkg-payments (service/) |
API + checkout flow; Stripe integration |
| Desktop clients | xkg-payments/desktop/ |
|
| Public API surface | Whatever is documented in docs/api-reference.md |
Vulnerability classes in scope: authentication and session management bypasses, authorization / privilege-escalation defects (including subscription-tier escalation), injection (SQL, command, template), unsafe deserialization, SSRF, path traversal, broken access control on user data objects, cross-origin data leakage that exposes another user's data, payment-flow manipulation (amount, SKU, entitlement), secrets or tokens committed to the repository, and the workers.dev-origin misconfiguration described in §2.3.
The following are not eligible for payout. We still want to hear about them — report them and we will fix them, but they do not carry a bounty, and reporting them will not be treated as a good-faith security report under this program.
| Out of scope | Rationale |
|---|---|
| Cloudflare Access bypass on the workers.dev hostname | This is a known, already-accepted configuration gap (§2.3). It is tracked and scheduled for remediation, and is not being disclosed for the first time through the bounty program. |
| Findings that depend on social engineering of a human | Out of scope regardless of severity. |
| Denial of service that requires sustained volumetric traffic | DoS via a small number of requests affecting other users remains in scope (§2.1). |
| Missing security headers with no demonstrated impact | Low/informational hardening; we will accept and act on these, without payout. |
| Self-XSS requiring the user to paste into their own console | Not a vulnerability. |
| Automated scanner output with no demonstrated exploit | Not a vulnerability. |
| Rate-limit exhaustion on a single user's own account | Not a vulnerability. |
| Reports about third-party services we do not control | Forward if a path exists; otherwise out of scope. |
| Source-code disclosure of our own proprietary code | See §7.4. |
| Findings in dependencies with no reachable exploit path | Report with the path; we will triage. |
This is documented deliberately and in advance. It is a known configuration state, not a hidden vulnerability.
access.api.error.not_enabled ("Access is not enabled").griptoad, the worker is reachable at
seele-memory.griptoad.workers.dev with no authentication layer.ORIGIN_URL, so a request
to the worker hostname can reach content that is also public elsewhere.seele-fresh, is not
deployed to the account (the account's script list contains only
seele-memory), but seele-fresh.griptoad.workers.dev remains a
resolvable public hostname returning Cloudflare error 1042
("worker not found"). A hostname that resolves is in scope to probe for
deployment mistakes even when it errors.Remediation is scheduled (§9, item U1) and is tracked independently of this bounty program. We are publishing it here so that:
Before U1 is complete, the worker hostname is a known public surface. Testing it is explicitly in scope for the following classes only:
Out-of-scope on the worker hostname: authentication bypass (there is no auth to bypass — that is the known issue), and "the hostname is public."
Payout amounts are fixed and are drawn from the pool described in §4. They do not scale with the pool balance, and the pool is never allowed to go negative (§4.4).
| Severity | Definition | Examples | Payout |
|---|---|---|---|
| Critical | Full compromise of the platform or of all user data. | Remote code execution; persistent authentication bypass; full account takeover; SQL injection yielding the full user table; a flaw that lets an unauthenticated attacker mint paid entitlements. | $500 |
| High | Significant unauthorized access to data or paid features. | Authorization bypass exposing another user's private notes; privilege escalation from Free to Pro without payment; payment-amount manipulation granting a paid tier; SSRF reaching internal metadata endpoints or other users' assets. | $200 |
| Medium | Real impact, but scoped or recoverable. | Cross-origin data leakage exposing non-public data of another user; denial of service affecting other users; stored XSS executing in another user's authenticated session; session fixation. | $75 |
| Low | Limited information disclosure with no direct user impact. | Disclosure of internal hostnames, version fingerprints, or stack details; enumeration of valid account identifiers; missing headers with no demonstrated impact. | $25 |
| Info | Not exploitable, but genuinely useful hardening. | A defensive-depth suggestion; a missing security header with no impact; a best-practice deviation. | $10 |
If you believe a finding is higher severity than we assign, reply to the original thread with your reasoning. A human will re-review. We do not require you to accept our initial classification to remain eligible.
Net Revenue for a calendar month M is:
Net Revenue(M) = Gross Collected(M) − Refunds(M) − Chargebacks(M)
− Taxes(M)
− Payment Processor Fees(M)
amount_received on all Stripe
PaymentIntents that reached status succeeded, plus the sum of the
first invoice amount on all Stripe Subscriptions that reached status
active in M. For subscriptions, the first charge only; recurring
renewals are attributed to the month they settle.M or any prior month. Refunds are not netted within the
month of the refund; they reduce the pool permanently.application_fee
and international card fees).Currency. All amounts are normalized to USD at the rate Stripe
reports on the balance transaction (balance_transaction.exchange_rate).
The ledger stores the USD amount and the original currency and rate for
audit.
Exclusions. The following never count as revenue and never fund the pool: taxes, processor fees, refunds, chargebacks, gift-card balances, credits issued outside a payment, and money that moves through the platform without a settled charge.
For each calendar month M:
Allocation(M) = 0.03 × Net Revenue(M)
The allocation accrues to the pool at the close of the following month. For example, the allocation for October 2026 accrues when October closes and is available for payout on 2026-11-01.
The pool is cumulative. Unspent allocations carry forward. There is no expiry and no clawback of unspent balance.
Pool Balance(M) = Pool Balance(M−1) + Allocation(M) − Payouts(M)
The pool ledger is a newline-delimited JSON file:
state/bounty-pool/ledger.jsonl
This file does not exist yet — it is created as part of the activation
checklist (§9, item R3). It will live in the xkg-payments repository, not
in the public website repository, because it contains financial figures.
Each line is one immutable event:
{"date":"2026-12-01T00:00:00Z","type":"allocation","month":"2026-11","gross_cents":0,"refunds_cents":0,"fees_cents":0,"net_cents":0,"allocation_cents":0,"pool_cents":0}
{"date":"2026-12-04T14:22:00Z","type":"payout","finding_id":"XKG-2026-0007","researcher":"<handle>","severity":"high","amount_cents":20000,"pool_cents":0}
Event types: allocation, payout, adjustment (manual correction, with a
mandatory reason field).
The public balance figure — total pool balance and lifetime paid out — is published on /security/acknowledgments/, which already refers to this ledger path. The public page shows the total only, not per-event detail.
If the pool balance is insufficient to cover an approved payout, the payout is deferred, not reduced and not refused. Deferred payouts are paid in full as soon as the pool has sufficient balance, in severity order, oldest first. We will notify the affected researcher.
A deficit can only arise if refunds or chargebacks exceed allocations. In that case the pool is held at $0 and resumes accruing from the next month's allocation.
The revenue calculation described in §4.1 is not implemented. As of 2026-11-14:
| Piece | Status |
|---|---|
xkg-payments service |
Not running. |
| Stripe API keys | Placeholders only (sk_test_replace, sk_live_replace). No live key. |
| Stripe webhook handling | Code exists (service/stripe_client.py::verify_webhook, service/checkout.py) but is not exercised, because no live Stripe account is attached and STRIPE_WEBHOOK_SECRET is a placeholder. |
| Revenue-calculation module | Does not exist. No code computes Net Revenue(M). |
| Ledger writer | Does not exist. |
Ledger file state/bounty-pool/ledger.jsonl |
Does not exist. |
| Monthly accrual job | Does not exist. |
This is the whole of what "not activated" means in practice. Until the table above is fully green, the pool balance is $0 and no payout can be made.
We commit to the following turnaround. "Acknowledged" means a human has read the report and sent a reply — not an autoresponder.
| Stage | Critical | High | Medium | Low | Info |
|---|---|---|---|---|---|
| Acknowledgment (human reply) | 24 hours | 3 business days | 3 business days | 7 business days | 7 business days |
| Triage (validity + severity decision) | 48 hours | 5 business days | 10 business days | 15 business days | 15 business days |
| Status update (if not yet fixed) | every 72h | every 7 days | every 14 days | every 30 days | as needed |
| Target fix | 7 days | 14 days | 30 days | 60 days | best effort |
| Disclosure window opened | 90 days after report | 90 days | 90 days | 90 days | 90 days |
If triage raises a report's severity, the SLA re-bases to the new severity from the moment of the change, and we will say so explicitly.
A payout is eligible when all of the following hold:
Payouts are made via a platform chosen at activation time. The platform and its minimum account and KYC requirements are user-owned — they are not something we can provision on the researcher's behalf. At the time of writing, the intended platform is:
| Requirement | Detail |
|---|---|
| Platform | To be confirmed at activation. Candidate: a hosted payout service (e.g. a marketplace-style platform with built-in tax handling) rather than manual wire transfers, to keep the user tax burden on the recipient rather than on us. |
| Account | The researcher must be able to create an account on that platform. Minimum: a valid email address and, depending on jurisdiction, a phone number. |
| KYC | Identity verification at the platform's own level. Typical minimum for an individual: government-issued photo ID + proof of address, or the platform's equivalent document check. |
| Tax form | The platform may require a W-8BEN (non-US) or W-9 (US) before the first payout. |
| Minimum payout | TBD at activation. If the platform's minimum exceeds the Low-tier payout, we will aggregate payments until the threshold is met rather than reduce the tier amount. |
| Currency | USD by default. Other currencies converted at the platform's payout-date rate, with the conversion disclosed on the ledger line. |
We cannot pay out without these. We will not ask a researcher for identity documents by email, and we will never ask for a password or a one-time code to "verify" a payout.
| Step | Target |
|---|---|
| Payout approved after triage | 5 business days |
| Payout processed after approval | Next payout run (weekly, once the platform is live) |
| Funds visible to researcher | Per platform; typically 1–3 business days after processing |
Report privately to:
Do not file a public issue, pull request, gist, blog post, or tweet.
Encrypted email to the address in security.txt is preferred; the PGP key
is at https://seele.agency/.well-known/pgp-key.txt.
Include, ideally:
Do not include: real user data you did not create, third-party personal data, bulk scraped data, a destructive proof of concept, or a persistent denial of service. We will not punish a researcher for an accidental overshoot; we will ask you to stop and clean up.
We want to be clear, because researchers often worry about this: we do not intend to use your report to bring a copyright or trade-secret action against you, and we will not hand your proof-of-concept code to law enforcement.
We also do not want to take your intellectual property. If you submit code that could be reusable in our codebase, you grant us a perpetual, worldwide, royalty-free, non-exclusive license to use it internally to fix the issue and in any future release. You keep all rights to your own code and may use it elsewhere, including publicly, once the disclosure window ends. We will credit you in the release notes where you want credit.
If your proof of concept is a tool you built for this, you keep it. We ask only that you not publish it before the coordinated-disclosure window closes.
After the window closes — or once we have fixed the issue and agreed a date — you are free to publish. We would like a heads-up so we can ship a patch and a changelog entry at the same time, but it is not a condition.
Seele Agency supports safe harbor for good-faith security research. If you comply with this policy, we will not pursue legal action against you for your research activity.
Safe harbor applies to your activity if all of the following are true:
Safe harbor does not apply to:
Safe harbor does not depend on, and is not conditional on, being paid. A report is eligible for safe harbor even if it is out of scope, even if it is a duplicate, even if it is rejected as invalid, and even if the pool is empty. Safe harbor is about how you behaved, not about whether we owe you money.
We cannot bind third parties. Safe harbor under this policy protects you from our claims. It does not protect you from claims by our hosting provider, our payment processor, or anyone else. We will make reasonable efforts to notify those parties of safe-harbor research when we are legally able to, and we will not voluntarily share your identity with them. If a third party does bring a claim despite our having made those efforts, we will tell you promptly and will not enforce our agreement against you to the extent doing so would increase your exposure.
We also cannot guarantee that safe harbor will be recognized in every jurisdiction or by every third party. We state it clearly anyway, because researchers need to be able to rely on something.
Split into what we did and what the user must do. This is the concrete list of everything required for the program to actually pay out.
| # | Item | Where | Status |
|---|---|---|---|
| D1 | Program policy written — scope, tiers, SLAs, disclosure, safe harbor, revenue calc | docs/bounty.md (this file) |
✅ |
| D2 | Public security policy page reflecting the bounty program | security/index.html |
✅ (pre-existing, 2026-08-01) |
| D3 | Public acknowledgments page with pool balance + severity table | security/acknowledgments/index.html |
✅ (pre-existing, 2026-08-01) |
| D4 | security.txt pointing at both |
.well-known/security.txt |
✅ (pre-existing, 2026-07-31) |
| D5 | Severity tiers + payout amounts defined | §3, mirrored in D2/D3 | ✅ |
| D6 | Response SLAs defined | §5 | ✅ |
| D7 | Safe-harbor statement written | §8 | ✅ |
| D8 | Revenue calculation defined | §4.1 | ✅ (spec only — see R4) |
| D9 | 3% allocation rule defined | §4.2 | ✅ (spec only) |
| D10 | Ledger format + path defined | §4.3 | ✅ (spec only) |
| D11 | Known workers.dev exposure documented as accepted risk, out of scope | §2.3 | ✅ |
| D12 | Activation checklist written | §9 | ✅ |
| D13 | Payout platform + KYC requirements identified | §6.2 | ✅ (identified; platform selection is U4) |
| # | Item | Why it blocks | Owner |
|---|---|---|---|
| U1 | Enable Cloudflare Access on the account, and bind an Access application (policy) to the seele-memory.griptoad.workers.dev hostname. Until this is done, §2.3 stands and the worker is unauthenticated. |
The known exposure in §2.3 is real. We document it, but the fix is a Cloudflare account change only the user can make. | User |
| U2 | Create a live Stripe account (or connect an existing one) and populate STRIPE_SECRET_KEY (live), STRIPE_WEBHOOK_SECRET, and STRIPE_API_VERSION in the xkg-payments environment. |
No live key ⇒ no revenue ⇒ no pool. Currently only placeholders. | User |
| U3 | Confirm the legal entity that receives the funds and enters the payout arrangement. | The contract with the payout platform and the tax position depend on the entity. | User |
| U4 | Choose the payout platform (hosted marketplace-style service vs. manual wire/ACH). | Determines KYC flow, minimum payout, and per-payout overhead. | User |
| U5 | Define the initial pool seeding policy — whether the pool starts at $0 on activation or is seeded with an opening balance. | §4.2 starts the pool at $0 by default; a seed would need to be an explicit allocation ledger event with a reason. |
User |
| # | Item | Depends on | Notes |
|---|---|---|---|
| R1 | Implement the revenue-calculation module in xkg-payments (Stripe balance-transaction / invoice aggregation per §4.1). |
U2 for live verification | Can be written and unit-tested against fixtures now. |
| R2 | Implement the ledger writer (state/bounty-pool/ledger.jsonl), the pool-balance computation, and the monthly accrual job. |
R1 | This is the artifact that makes "3% of revenue flowing" true. |
| R3 | Create the ledger directory + initial ledger.jsonl (possibly empty) so the path referenced by /security/acknowledgments/ exists. |
— | Do this now — the public page already links to it. |
| R4 | Add a reconciliation test that recomputes pool balance from the ledger and asserts it equals the published figure on /security/acknowledgments/. |
R2 | Prevents the public number drifting from the ledger. |
| R5 | Publish the live pool balance on /security/acknowledgments/ from the ledger instead of hardcoding $0. |
R2, R3 | Today the page hardcodes $0; make it data-driven so it stays truthful as revenue arrives. |
| R6 | Implement the payout-run job: select approved findings, check pool sufficiency, emit payout ledger events, and generate researcher payment instructions. |
U4 | §4.4 deferral logic belongs here. |
| R7 | Add a webhook → ledger path so refunds and chargebacks reduce the pool (§4.1). | U2, R1 | Without this the pool can silently drift upward. |
| R8 | Add a publish step that regenerates the pool-balance figure on /security/acknowledgments/ monthly. |
R5 | Matches the "we publish the total here monthly" claim on that page. |
The bounty program is activated when all of the following are true:
/security/acknowledgments/ is computed from the ledger and matches it (R4 passes).| Version | Date | Change |
|---|---|---|
| 1.0 | 2026-11-14 | Initial program. Policy, tiers, SLAs, safe harbor, revenue calculation, activation checklist. |
Substantive changes are announced on /security/acknowledgments/ at least 14 days before they take effect, except changes that reduce a payout tier or narrow scope, which take effect immediately for reports already submitted and accepted.
| Purpose | Address |
|---|---|
| Security reports | security@seele.agency |
| PGP public key | https://seele.agency/.well-known/pgp-key.txt |
| Policy questions | security@seele.agency |
| Bounty payout questions | security@seele.agency |
| Acknowledgments / credit requests | security@seele.agency |
This program is a design and instrumentation commitment. The policy is live; the funding mechanism is pending. See §0 and §9.