Service Policy

1. AI processing disclosure

• Agent Kits Lab uses AI-assisted tools, including Hermes and the engagement-specific model/API provider(s) named in the scope receipt, to inspect authorized material, classify findings, draft reports, and perform QA.
• AI output is untrusted draft evidence until checked against repository files, deterministic command output, or cited sources.
• Customer material is not intentionally used by Agent Kits Lab to train a public model. Third-party handling remains subject to the named provider/account terms.
• Confidential source is not intentionally pasted wholesale into remote prompts; excerpts, summaries, names, hashes, and local deterministic tools are preferred.
• Before payment, the scope receipt states LOCAL_STATIC_ONLY or names every material remote AI/API provider expected to receive non-public content. A generic label is insufficient.
• If provider selection changes, processing stops until the customer approves the revised provider or elects local/static-only handling.

2. Authorization and data minimization

• Only repositories/artifacts the customer owns or is authorized to provide may be audited.
• Intake excludes production credentials, .env values, private keys, tokens, browser profiles, customer databases, personal data, and unrelated proprietary material.
• Agent Kits Lab does not bypass access controls or seek broader permissions merely because credentials are available.
• Accidentally encountered secrets are redacted and described by location/type, not value.

3. Execution and isolation

• Default mode is static, read-only inspection.
• Customer build/test/lint/install scripts are not executed unless the accepted scope permits named commands and a suitable isolated environment is available.
• Isolation requires a disposable container/VM or equivalent; no inherited credentials; network off by default; bounded CPU, memory, disk, process count, and timeout; explicit output directory.
• If safe isolation cannot be established, commands are NOT_RUN. Missing evidence is not relabeled as PASS.
• Base-audit acceptance never authorizes repository writes, production deployment, destructive actions, or credential use.

4. Retention and deletion

• Early deletion may be requested, subject to legal/payment/security preservation.
• Deletion means removal from service-controlled active storage; instantaneous erasure from third-party backups/provider logs is not promised.

5. Confidentiality boundary

• Non-public code, architecture, findings, and business context are confidential and used only for accepted scope.
• Access is limited to the operator, named AI/tool providers, and necessary service infrastructure.
• Reports or customer identity are not used as samples, testimonials, training data, or marketing proof without separate permission.
• Exceptions include already-public information, independent development, lawful unrestricted receipt, or legally compelled disclosure.

6. Security incidents

1. Stop processing and contain access.
2. Preserve minimum evidence.
3. Notify the customer promptly with known scope and unknowns.
4. Recommend credential rotation where relevant.
5. Document remediation and deletion.
6. Do not publicly disclose customer details without permission unless required.

No absolute security guarantee is offered.

7. Delivery, corrections, and revisions

• Target delivery is within 5 business days after cleared payment and complete accepted intake. If capacity/scope cannot meet it, payment is not requested until revised terms are agreed.
• One clarification round is included within 7 calendar days when it requires no new repository version or analysis.
• Confirmed factual evidence errors reported within 14 calendar days are corrected without expanding scope.
• New commits, repositories, implementation, or materially changed requirements are new scope.
• Revised reports carry a version/date and change note.

8. Disputes, refunds, and limitations

• The parties first compare accepted scope, snapshot/commit, cited evidence, payment record, and delivered artifacts.
• If a deliverable is materially missing, Agent Kits Lab may complete it, correct it, or recommend an appropriate refund under accepted checkout terms.
• Disagreement with a rating, absent business outcomes, or later repository changes does not itself establish non-delivery.
• Payment/refund execution and custom legal/liability commitments require human approval.
• The audit is point-in-time engineering review—not legal advice, penetration testing, certification, warranty, or a guarantee of security, productivity, revenue, compatibility, or defect-free software.

9. Material service providers and changes

• Payment/commerce: Ko-fi and the connected processor process checkout/payment data. Final checkout displays the actual processor.
• Email: Cloudflare Email Routing forwards role-address mail to the operating mailbox; its provider processes the forwarded message.
• Remote AI/API: engagement-specific. Scope names each material provider before payment or states LOCAL_STATIC_ONLY.
• Material policy changes apply prospectively with a dated version; accepted work follows its incorporated version unless both parties agree or law/security requires a change.

10. Acceptance record

Intake is complete only when it records requester/authorization, exact snapshot/exclusions, execution permission, remote-AI mode/providers, retention exception, policy version, recipient, Agent Kits Lab scope acceptance, private checkout, and cleared-payment receipt.