mayim-agent 2.0.0__tar.gz

mayim_agent-2.0.0/CHANGELOG.md

@@ -0,0 +1,554 @@
+# Changelog
+
+## 1.18.15 — candidate
+
+- Bump the distribution and MCP package metadata to 1.18.15; align current
+  candidate documentation and packaging checks with that version.
+- Retain the S3 SDK code, dependency requirements, entry points and fixed
+  contract resources. Publication remains a separate Release step.
+
+## 1.18.14 — candidate
+
+- S3: encrypt the v3 journal with AES-256-GCM and a separately provisioned key;
+  preserve the original Claim/alias/report across process restart and key failure.
+  Validate and atomically encrypt pre-publication S2 journals on explicit reuse.
+  Keep all ten responder operations and the fixed B4/legacy pack bytes.
+
+- Add StructuredResponseClient, typed local answer validation, private durable
+  journal, explicit ten Python/MCP operations and same-ID/bytes recovery.
+- Keep profile omission/unknown/non-disclosure and independent status states.
+- Identify Choice 2.1.0 by its separate v2_1 namespace; retain old pack bytes.
+- Update the B4 requester contract/guide/manifest identity and supported v3 digest;
+  retain the accepted S1 responder behavior and unchanged Choice2.1 bytes.
+- Mirror the DC-fixed B4 v3/Choice2.1 producer packs; align closed DTOs and public
+  terms digests. Keep B aliases and withheld sources through same-byte recovery.
+- Add EN/JP respondent Guide and isolated Python/MCP/wheel integration evidence.
+  Live provider/wallet/OpenClaw checks and publication remain Release work.
+
+## 1.18.9 — candidate
+
+- URL Choice & Reason strict worker DTOs, private Claim/answer recovery, Requester paid registration/read recovery and result proofs.
+- Immediate v2 role UA dispatch with explicit v1 compatibility.
+- Public positioning, memory-only access recovery and paid POST terms-check explanation.
+- Two byte-identical Hondo packs and isolated new/legacy client conformance.
+- Candidate qualification does not assert Development Control acceptance or public release.
+
+## 1.18.8
+
+- Add opt-in Base mainnet access-quota purchases for all four Task families,
+  with a shared access-only budget; automatic spending remains disabled by default.
+- Preserve the original request and payment proof while a purchase is pending,
+  count confirmed payments once, and retain purchased credits after result expiry.
+- Allow a new explicit GET after terminal result recovery and an explicitly
+  selected new purchase after a Paid Claim access purchase definitively fails.
+- Preserve Claim safety, readonly recovery, transport deadlines, and the keyless,
+  inspect-only MCP scope. Access purchase state remains memory-only.
+- Include 1.18.8 package/MCP metadata and exclude private evidence and operational
+  material from distributions. The GitHub Release is published; a fresh live-payment
+  retest of the c4 documentation artifacts and independent PyPI/MCP Registry readback
+  are not asserted here.
+- Details: [1.18.8 notes](docs/release_notes/v1.18.8.md) and
+  [access quota usage](docs/access-quota.md).
+
+## 1.18.7
+
+- Retain Payment-Required and Payment-Response through standard Paid HTTP,
+  including case variants and duplicate values, without changing other fetchers.
+- Preserve actual received HTTP status and validated LN error code/reason/request
+  ID across safe exceptions and same-request Claim recovery. Read 1.18.6 records.
+- Explain recovery before an Execution ID or credential is returned, using the
+  original Task/key/version/private directory across process restarts.
+- Keep fixed HTTP limits, contract packs, once-only purchases and report identity.
+- Let standard Paid LN API reads use the remaining overall request budget,
+  preserving its finite deadline, connection bounds and zero automatic retries.
+
+## 1.18.6
+
+- Persist Paid Service Trial Claim requests before sending and provide explicit
+  same-key recovery across restarts and response/storage failures. Return usable
+  credentials only after the private credential and purchase journal are saved.
+- Accept unrelated provider discovery extensions after known-condition checks;
+  preserve seller address spelling and existing once-only purchase/report rules.
+- Keep API/Task Definition versions and both formal contract packs unchanged.
+
+
+All notable changes to the `ln-church-agent` SDK will be documented in this file. Detailed release notes for specific versions can be found in the `docs/release_notes/` directory.
+
+## [1.18.3] - 2026-09-14 (Private Source Work — Immediate HTTP Visit)
+
+* Adds the typed `AgentImmediateVisitClient`, `ImmediateVisitExecutor`, `FrozenImmediateVisitReport`, and private `ImmediateVisitJournal` for explicit `immediate_http_visit.v1` work.
+* Selects one immutable endpoint after Claim, performs one bounded GET, automatically derives the fixed UTF-8 HTML/JSON fingerprints, and records comparable or reasoned inconclusive results without buying target HTTP 402 responses.
+* Preserves GET start and Report identity across crashes. Recovery uses authenticated status and the same canonical Report; Completion allows at most three POSTs and explicit polling defaults to five requests one second apart.
+* Separates the 48-hour listing, 10-minute first-Report deadline, accepted-Report recovery, server-controlled re-entry, evaluation, and payment. Base reward is 0.0075 USDC with a possible 0.0075 USDC exact-match bonus.
+* Documents v1.17 C50/C500/C5000 capacities of 50/500/5,000 for 1/10/100 USDC while retaining its 0.01 USDC reward and existing contracts.
+* Packages the isolated JustHTML vendor, license and compatibility provenance, and immediate-visit resources without changing Python `>=3.8.1`, extras, entry points, or inspect-only MCP authority.
+* Provides English API examples, official Worker/Requester Guide destinations, and free Task Board result links. Private source work does not establish source acceptance, live publication, deployment, or payout.
+* Details: [v1.18.3 notes](docs/release_notes/v1.18.3.md).
+
+## [1.18.0] - 2026-08-15 (Private Source Candidate — Scheduled HTTP GET Batch SDK)
+* **Generic v2 Task profile**: Adds strict v2 Task detection, one-attempt Claim, readiness, abandonment, compound Completion, and Submission-status recovery for `scheduled_http_get_batch.v1`, while preserving released v1 models, flat errors, routes, and retry semantics. An indeterminate Claim becomes `CLAIM_OUTCOME_UNKNOWN` and is never replayed.
+* **Controlled Manifest transport**: Reuses the exact signed Manifest URL, resolves fresh A/AAAA answers per attempt, and validates IPv6 fail-closed against a pinned, runtime-network-free IANA allocation and special-purpose snapshot. It rejects a whole answer set containing any forbidden address before socket creation, pins the numeric peer, verifies connected-peer equality, and preserves canonical-host TLS SNI, certificate, and `Host` handling. Redirects, ambient proxies, cookies, authorization, `netrc`, request bodies, caller headers, and automatic decoding are prohibited.
+* **Finite fetch policy**: Allows at most three Manifest attempts within five seconds, limits response headers and the successful body to 32 KiB, reads zero non-200 body bytes, and applies encoding or network-policy rejection before an otherwise retryable status. Signed query values are redacted and never forwarded.
+* **Sequential one-attempt targets**: Runs targets in Manifest order, durably writes `ATTEMPT_STARTED` before target I/O, never retries or resends an indeterminate target, and records typed unattempted results when a new target cannot start before T+5.
+* **Crash-safe local journal**: Binds a versioned, checksummed journal irreversibly at genesis to the Task ID, full Task Definition tuple, one-way local Claim handle, and derived Execution ID, then binds the Manifest digest durably at T. Same-directory atomic replacement, flush, lock, native-Windows close-before-replace, and supported directory durability fail closed on missing, legacy-unbound, definition-mismatched, corrupt, or ambiguous state without storing the Claim token or signed URL.
+* **Exact Completion recovery**: Freezes canonical report bytes and serializes Completion recovery under a stable sibling cross-process operation guard. Recovery is status-first even with zero prior dispatch attempts; only a fresh Completion may reserve attempt one, and only one absent-status decision before close may permit attempt two with the same Submission ID and bytes. It recovers an existing on-time receipt after close and treats `COMPOUND_COMPLETION_ACKED` as the terminal dispatch boundary. A later authoritative bound `terminal=true` status is atomically retained as `TERMINAL_STATUS`; nonterminal or mismatched status evidence cannot advance the journal.
+* **Integration boundary**: Adds a public-safe CLI/example and public exports without adding Task mutation to the keyless inspect-only MCP. Claim creates the required definition-bound journal before making its credential usable; run/resume never recreates a missing journal. Signed URL state is private, slotted, non-serializable, and redacted. The canonical contract fixture is packaged byte-for-byte in wheels and sdists. No dependency is added.
+* **Qualification boundary**: Linux candidate evidence includes the fixture-driven exact 42-row production-connector network matrix. Native macOS and native Windows filesystem, locking, atomic-replace, DNS, socket, and TLS qualification remain separate required lanes. Mocked platform branches do not qualify a native platform, and native Windows qualification does not require PowerShell 5.1.
+* **Candidate status**: This Private source candidate is pending independent audit and does not claim release readiness, Public promotion, publication, deployment, server acceptance, Evaluation acceptance, reward approval, or payout.
+* **Details**: [v1.18.0 candidate notes](docs/release_notes/v1.18.0.md)
+
+## [1.17.1] - 2026-08-11 (Private Integrated Candidate — Reward Destination Education and Agent-Earning Documentation)
+* **Reward destination education**: Documents that Public Agent Task execution and observation require no wallet credential, while a reward-bearing Claim requires a non-zero EVM `reward_address` for USDC on Base (`eip155:8453`). The address is fixed in the immutable Claim snapshot, cannot be changed later, and is not control-verified by Hondo.
+* **Secret and destination boundary**: The Task CLI and documentation never request, store, or send a wallet secret, private key, seed phrase, signer credential, or wallet-control proof. Lightning, LNURL, and BOLT11 remain unsupported as Task reward destinations.
+* **Agent-earning front**: Adds the exact first-job/first-reward messages and the canonical `Discover → Read Definition → Claim → Execute → Register Observations → Complete → Check Evaluation → Check Reward` lifecycle, while keeping execution with the Host Agent and preserving the SDK's no-crawl, no-browser, no-login, no-scan, and no-target-payment boundary.
+* **State and authority separation**: Keeps the Hondo response and immutable Claim snapshot authoritative. Claim success, Completion receipt, Evaluation, reward approval, and payout remain separate; `accepted` is not `paid`, Task availability and income are not guaranteed, and eligible rewards remain best-effort with finite retry rather than a payment SLA.
+* **Compatibility**: JSON, wire, and Claim request payloads are unchanged. Education is limited to README, CLI help, and human-readable Task output; public API, schemas, transport, retry, checkpoint, credential, payment, Evaluation, and settlement behavior remain unchanged.
+* **Version materialization**: Updates package, User-Agent, and MCP Observation identities consistently to `1.17.1` without changing dependencies.
+* **Known environment boundary**: Windows plus Python 3.14 remains unsupported because of the transitive `coincurve` compatibility boundary; Python 3.11 remains recommended on Windows.
+* **Candidate status**: This Private integrated candidate does not implement a Hondo runtime change and does not claim Public promotion, release, publication, or deployment.
+* **Details**: [v1.17.1 candidate notes](docs/release_notes/v1.17.1.md)
+
+## [1.17.0] - 2026-07-28 (Private Source Candidate — Agent Task Venue SDK)
+* **Public Agent Task lifecycle**: Adds a synchronous, wallet-keyless `AgentTaskClient` for discovery, detail, Claim, public-safe Observation Register, Completion, and bounded reward-status polling for the domainless `payment_surface_discovery.v1` Task at the fixed LN Church origin.
+* **Interface authority and 1:N cardinality**: Treats the public Hondo API as canonical and the SDK as an optional supporting client. One Task record is one Task Offer with Hondo-defined homogeneous capacity, while every successful Claim creates an independent Execution.
+* **Server-owned Offer snapshots**: Strictly retains Hondo-provided execution, Claim, reward-paid, capacity, `claimable`, and `poc_terms` fields. `OPEN` is an Offer publication status—not proof of claimability—and the SDK never reconstructs aggregates, predicts Claim success, or substitutes local time/capacity inference for the server snapshot.
+* **Hondo policy boundary**: Displays server-provided registration, reward, maximum-principal, and capacity values without hardcoding, deriving, decrementing, reserving, or re-aggregating them. Alpha A/B/N values are contract evidence and Hondo policy, not SDK calculation rules; generic canonical atomic amounts and internally consistent capacities remain valid.
+* **No-payment transport**: Uses a dedicated fixed-route transport with public-IP DNS validation and pinning, preserved TLS hostname verification, bounded requests and responses, no redirects or ambient credentials, and fail-closed handling for unexpected HTTP 402 responses.
+* **Claim credential safety**: Keeps the lease-bound claim token out of normal model serialization, logs, exceptions, and CLI output; the CLI uses create-exclusive private credential files and secret-free ambiguity tombstones.
+* **Immutable Claim binding**: Retains the API-supplied Task Definition reference as public immutable metadata, never executable instructions, and binds each independent Claim credential to its authoritative Claim-time reward snapshot. The reward address remains only a payout destination, never the Agent principal or wallet-control proof.
+* **Host-Agent discovery boundary**: Keeps open-web discovery with the Host Agent or another third-party runtime. The Task has no Task-level domain or parameters and carries empty `seed_urls`; the SDK neither crawls nor searches the target and adds no browser, JavaScript, login, form, scan, payment, mutation, or Task Offer registration capability.
+* **Public-safe Observation contract**: Adds strict bounded schemas and stable RFC 8785 JCS submission identity. An actual HTTP 402 discovered surface is required, but subtype precision is not an eligibility gate, so an `unknown` 402 remains valid. Repeated domains, URLs, or known surfaces are not locally inferred to be reuse or failure.
+* **Deterministic guided bridge**: Makes `submit_and_complete_domain_observation()` and `task submit-complete` the standard path from a validated Observation through Register and Completion. Completion IDs are derived only from the strictly verified Register receipt; callers neither interpret nor re-enter them.
+* **Secret-free restart checkpoints**: Persists `REGISTER_PENDING` before Register and `REGISTERED` before Completion through a durable sink. Resume reuses the exact submission identity or skips an already verified Register, while the checkpoint remains non-authoritative restart metadata with no Claim token or usable credential.
+* **Separated status scopes**: Keeps Offer `OPEN`, Claim-response `CLAIMED`, and the six public Submission statuses distinct; private pre-submission `EXPIRED` and `ABANDONED` states never validate on a public Submission resource.
+* **Exact reward outcomes**: Maps public Submission status to the six reward states (`pending`, `not_eligible`, `approved_pending`, `paid`, `failed`, and `ambiguous`) and accepts only the finite status-specific failure codes, including `claim_task_or_observation_binding_mismatch`. Free-form, cross-status, and otherwise inconsistent outcome combinations fail closed.
+* **Claim-specific result semantics**: Uses the existing Task, Submission, and Observation IDs plus the successful Claim's immutable Definition and reward snapshots for status, without Task-level domain equality. A possibly stale TaskGet advertised reward is never Claim authority, and the Definition remains metadata that the SDK neither downloads nor executes.
+* **Task-detail Execution pages**: Retains Hondo-provided public-safe Execution summaries only on Task detail, paginated through that route's existing `limit` and opaque `cursor`. List items never acquire summaries, and the SDK neither reaggregates a page nor invents an Execution identifier.
+* **Network-free Claim verification**: Provides local verification that binds a public Execution summary to a known validated Claim snapshot and its existing public identities. Structure-only Task-detail parsing is not presented as Claim-bound proof, and the verifier never uses TaskGet advertised reward or reevaluates Observation quality.
+* **Durable Completion and bounded polling**: Completion 2xx records durable Submission/CompletionReport acceptance only—not Evaluation acceptance, Settlement, or payment. Bounded polling returns the last valid `pending` or `approved_pending` result as non-terminal when its attempt, exchange, or observation window ends.
+* **Low-level compatibility**: Keeps the existing DomainObservation Register/response and internal OpenClaw interfaces for manual-ID integrations, but removes the old public Task type and Task Offer models rather than adding a compatibility alias or paid wrapper.
+* **PoC reward disclosure**: Preserves Hondo's canonical disclosure that Completion is durable receipt only and reward sending is automatic best-effort with finite retry and recorded evidence, without promising a payment SLA, individual investigation, manual resend, compensation, alternative payment, or arbitrary non-payment.
+* **Compatibility boundary**: Preserves the internal OpenClaw endpoints and `X-Internal-Secret`, existing payment-capable `LnChurchClient`, and inspect-only MCP without adding Task mutation tools.
+* **Candidate status**: This Private Source Candidate is pending independent audit and does not claim cross-repository compatibility, runtime acceptance, release readiness, deployment, or publication.
+* **Details**: [v1.17.0 candidate notes](docs/release_notes/v1.17.0.md)
+
+## [1.16.4] - 2026-07-19 (Private Candidate — Inspect MCP SSRF and Privacy Boundary)
+* **SSRF boundary**: Restricts Inspect to public HTTP(S) `GET`/`HEAD`, applies one shared hostname/address policy before DNS and to every resolved address, pins connections to vetted IPs, and revalidates every redirect hop and explicit Observation target.
+* **Public identity and privacy**: Publishes only the canonical origin of the initially requested URL. Redirect destinations remain transport-only, forbidden/internal URLs are wholly redacted, and attacker-controlled public scalars use fixed values or finite allowlists.
+* **Response and deadline limits**: Requests identity encoding, rejects content-encoded responses before body read, caps raw response material at 1 MiB, and applies one total wall-clock deadline across DNS, connect/TLS, status/header, body, and redirects.
+* **Observation safety**: Keeps submission explicit and canonical, sends an Observation POST at most once, never reads its response body, and reports ambiguous delivery as `observation_delivery_unknown` without replay.
+* **Failure contract**: Keeps response adaptation, parsing/classification, and network transport in distinct failure domains. Known invalid challenges are rejected without execution; unexpected parser faults stop safely.
+* **Release boundary**: This is a Private candidate pending independent re-audit. It does not claim formal Go, Public promotion, tagging, publication, deployment, or registry release.
+* **Details**: [v1.16.4 candidate notes](docs/release_notes/v1.16.4.md)
+
+## [1.16.3] - 2026-07-18 (P0-2 Canonical Payment and Privacy Boundaries)
+* **Canonical boundary**: Binds approvals and generated credentials to one canonical requirement, final wire URL (including GET query), request identity, idempotency key, expiry, and requirement hash; revalidates the approved snapshot before signing or sending.
+* **Settlement safety**: Separates credential generation, signature verification, and settlement verification; records unknown delivery outcomes as `settlement_unknown`; and makes session-budget check/reserve/confirm/release transitions atomic and operation-specific.
+* **SVM fail-closed**: Stops canonical SVM exact high-level sync and async auto-payment after approval and before session-budget reservation, signer, fee-payer or credential access, Solana RPC, credential generation, or a second HTTP request.
+* **SVM payload interoperability**: Keeps the low-level payload builder and validator available for payload construction only, using `SetComputeUnitLimit → SetComputeUnitPrice → TransferChecked → exactly one Memo`; Memo behavior was independently checked against the x402 Python 2.16.0 facilitator verifier with and without `extra.memo`.
+* **Privacy and navigation**: Redacts credential-bearing headers, payload keys, URL queries, receipt tokens, signatures, macaroons, and preimages at persistence and reporting boundaries; strips `Signature`, `Signature-Input`, and `DPoP` across untrusted redirects.
+* **Release boundary**: This entry records previously audited P0-2 behavior in a Private release-preparation state; this new release SHA remains pending independent re-audit. Public promotion, tagging, PyPI upload, and compatibility claims beyond the tested dependency set remain pending.
+
+## [1.16.2] - 2026-07-13 (Buyer-side Payment Boundary Hardening)
+* **Fixed**: Preserves canonical atomic payment requirements from parsing through policy evaluation, signer input, and signer-output verification without rebuilding exact amounts from floats.
+* **Hardened**: Requires `bolt11>=2.1.0,<2.1.1` for signed Lightning invoice validation and Windows-wheel-compatible `coincurve` support without a native `secp256k1` build; rejects invalid, amountless, placeholder, or mismatched L402/MPP invoices before wallet invocation; and treats MPP auth-parameter names case-insensitively while rejecting empty, duplicate, falsey, non-canonical, or contradictory amount/currency/invoice declarations.
+* **Fixed**: Rejects incomplete, invalid, or contradictory `Payment request` drafts before wallet, delegate, signer, retry, Evidence-success, or budget side effects regardless of the legacy fallback flag; only a complete draft retains the existing explicit opt-in fallback behavior.
+* **Hardened**: Validates trusted EIP-3009 domains and authorization signatures, and validates SVM `TransferChecked` instructions plus Compute Budget bounds of 200,000 CU, 1,000,000 micro-lamports/CU, and 200,000 lamports total priority fee.
+* **Security**: Applies case-insensitive strict netloc policy matching, removes camel/snake/hyphen credential families from nested navigation payloads and Evidence submissions, and discards original payload data on cross-origin redirect or HATEOAS navigation.
+* **Safety**: Validates EVM/SVM signer output before marking an operation irreversible, carries one Idempotency-Key-aware fingerprint across paid retries and HATEOAS navigation, reserves a canonical ambiguous amount once only after an irreversible result is lost, and keeps Evidence repository failures secondary to the primary payment outcome.
+* **Compatibility**: Preserves compatible pre-irreversible exception types and direct exception bases with secret-free replacement messages, retains v1.16.1 public constructors, execution signatures, the six-field `ExecutionContext` serialization shape, and Protocol member sets, keeps CLI inspection annotations evaluable on Python 3.8, and declares Python `>=3.8.1`.
+
+## [1.16.1] - 2026-07-04 (Verified Domain Track Semantics & Standards Alignment)
+* **Changed**: Renames the public framing toward "Domain-Control Verified Observation Track Lite" while keeping `verified_domain_track_lite` as a backward-compatible plan ID.
+* **Changed**: Treats `domain_control_verified` as the primary public semantic. Keeps `domain_owner_verified` only as a legacy compatibility field.
+* **Clarified**: Explicitly documents that `/.well-known/ln-church-domain-sponsor.json` is an LN Church domain-control challenge document, not an AI discovery standard.
+* **Added**: Adds standard-adapter metadata for future `/.well-known/ai`, A2A Agent Card, AI Catalog, and DNS TXT challenge support.
+* **Added**: Adds `/.well-known/agent-card.json` to OpenClaw discovery seed candidates.
+* **Safety**: Preserves all safety boundaries: not legal ownership proof, not recommendation, not verdict, not security scan, not certification, not trust score.
+
+## [1.16.0] - 2026-07-03 (Verified Domain Track Lite)
+* Added SDK and CLI support for Verified Domain Track Lite.
+* Added `register_verified_domain_track()`, `get_verified_domain_track_status()`, `get_domain_verified_track()`, and `save_verified_domain_track_proof()`.
+* Added `observe-domain track register/status/domain`.
+* Added `--proof-file` support for sponsor challenge / verify.
+* Safety: Verified Domain Track is not legal ownership proof, not a certification, not a recommendation, not a security scan, and not a trust score.
+
+## [1.15.0] - 2026-06-28 (Verified Domain Sponsor MVP v1)
+* **Added**: SDK helpers for Verified Domain Sponsor (`create_domain_sponsor_challenge()`, `verify_domain_sponsor()`, `save_domain_sponsor_challenge_document()`).
+* **Added**: CLI commands `ln-church-agent observe-domain sponsor challenge` and `ln-church-agent observe-domain sponsor verify`.
+* **Added**: Public-safe models for sponsor verification read models (`DomainSponsorVerification`, `DomainSponsorVerificationSummary`, etc).
+* **Safety**: Verified Domain Sponsor proves only domain-control challenge publication. It is not legal ownership proof, not a certification, not a recommendation, not a trust score, and not a security scan.
+* **Safety**: SDK does not automatically issue challenges or verify sponsors after paid registration. CLI avoids printing proof headers and challenge tokens unless explicitly requested with `--json` or `--print-document`.
+
+## [1.14.2] - 2026-06-27 (MCP Registry Metadata Fix)
+* **Fixed**: Shortened the `server.json` description to satisfy the official MCP Registry `description.length <= 100` validation rule.
+* **Changed**: Preserved the inspect-only MCP positioning in a more compact registry-safe form: keyless, no wallet, no signing, and no payment execution.
+* **Safety**: No runtime behavior, payment execution logic, settlement serialization, telemetry submission, or API surface was altered.
+
+## [1.14.1] - 2026-06-27 (Runtime Mode Boundary Clarification)
+* **Documentation**: Explicitly separated the SDK's positioning into two distinct modes: **Inspect-only Mode** (keyless, safe for enterprise preflight) and **Execution Runtime Mode** (policy-aware settlement loop).
+* **Added**: Expanded the Capability Matrix (`docs/07_capability_matrix.md`) and the `get_capability_matrix()` helper with new strict boundary fields (`mode`, `requires_private_key`, `can_execute_payment`, `can_submit_telemetry`, `auto_submits_telemetry`).
+* **Changed**: Updated `ln-church-agent-mcp` metadata to explicitly declare its keyless, inspect-only nature.
+* **Safety**: Added rigorous unit tests to ensure the MCP inspect entrypoint never calls the payment execution engine. No payment execution behavior or auto-telemetry logic was altered.
+
+## [1.14.0] - 2026-06-23 (Paid Domain Observation Slot)
+* **Added**: Introduced `register_domain_observation_slot` to purchase a 7-day public-safe observation run for a specified domain (x402 Paid Action).
+* **Added**: Introduced public read models `get_domain_observation_request` and `get_domain_observation_read_model` to safely query the observation status and discovered surfaces.
+* **Added**: Introduced internal observer APIs (`claim_domain_observation_targets`, `submit_domain_observation_result`) for observatory workers, strictly protected by `X-Internal-Secret`.
+* **Added**: CLI commands `observe-domain` for paid registration/status, and `observatory` for internal worker operations.
+* **Added**: Robust domain validation (`validate_public_domain_for_observation`) to prevent SSRF (blocking IPs, localhost, metadata endpoints).
+* **Security**: Enforced strict `public-safe` constraints (GET/HEAD only, no forms, no login, no payment to target).
+* **Architecture**: Clear separation between `requester_paid` and `domain_owner_verified`. The slot purchase does not imply domain ownership, endorsement, security scan, or certification (`not_a_verdict: true`).
+
+## [1.13.0] - 2026-06-14 (Observation Provenance / Protocol Roles / Verification Cost Vector)
+* Added public-safe `observation_provenance` metadata helpers.
+* Added `protocol_roles` shape for role → protocol → capability observations.
+* Added optional `verification_cost_vector` helper for sdk-reported, externally checkable cost counters.
+* Added optional `protocol_roles` and `verification_cost_vector` fields to explicit observation submissions.
+* Maintained v1.12.0 Reporter Identity Verification boundaries: no mandatory verification, no automatic hooks, no trust scoring, no recommendation/ranking.
+
+## [Backend Compatibility Note] - 2026-06-08 (Reporter Identity Boundary Hardening)
+* **Backend Update**: The SDK package remains `v1.12.0`. This update adds tests and documentation to verify compatibility with the latest Hon-den backend hardening.
+* **Backend Hardening**: Explicitly persists `ReporterAgentId` across all telemetry metadata. Distinguishes `self_reported` from `unknown` when AgentProfiles lookups fail.
+* **Backend Hardening**: Adds machine-readable safety boundaries (`verification_semantics="key_control_only"`, `not_a_trust_score=true`) to identity verify responses.
+* **Docs/Tests**: Added regression tests ensuring `ensure_reporter_verification()` safely handles the new response flags without requiring a new SDK release.
+
+## [1.12.0] - 2026-06-07 (Reporter Identity Verification Layer)
+* **Added**: Introduced an optional, client-managed **Verifiable Reporter Identity Layer** to explicitly prove private key control behind an `agentId`.
+* **Added**: Added `LnChurchClient.ensure_reporter_verification()` and `ensure_reporter_verification_async()` to automatically handle the full identity challenge-signature-verification lifecycle with in-memory caching.
+* **Added**: Embedded reporter verification metadata inside all core telemetry frameworks (`GoalAttemptIngest`, `ExternalObserve`, and `FailureObserve`) to allow downstream registries to distinguish between anonymous and cryptographically verified claims.
+* **Added**: Upgraded the Graph Ingestion engine (`AgentGraphSync`) to dynamically project `IdentityProof` nodes linked to the target agent via `HAS_IDENTITY_PROOF` edges.
+* **Added**: Enriched S3 compact read models with `reporter_verification_mix` (inside `goal-attempt-summary.json`) and `verified_reporter_attempt_count` (inside `goal-surface-candidates.json`).
+* **Safety**: Enforced strict non-mandatory boundaries; `self_reported` telemetry remains 100% functional and accepted. The layer proves key-control only, **never report truth or correctness**, and explicitly bypasses automated execution loops to prevent token drainage.
+
+## [1.11.3] - 2026-06-03 (Experimental Intent Signature Observation Sidecar)
+* **Added**: Experimental `intent_signature` and `classification_claims` sidecars for explicit Goal Attempt observations.
+* **Safety**: No payment execution behavior changed. No automatic telemetry hooks added. Not a recommendation or stable taxonomy.
+
+## [1.11.2] - 2026-06-03 (Grant-like Signal Detection Sidecar)
+* **Added**: Inspect-only Grant-like signal detection appended as a sidecar to `InspectResult`.
+* **Added**: Exposes `grant_signals` locally in the MCP `inspect_paid_surface` output.
+* **Safety**: No payment execution behavior changed. No grant redemption, no marketplace, no auto-submission to Hon-den.
+
+## [1.11.1] - 2026-05-31 (Auth-Capture Inspect-Only Standards Alignment)
+* **Added**: Inspect-only classification for x402 `auth-capture` mapping it strictly to the x402 settlement rail.
+* **Safety**: Absolute execution guard for `auth-capture` in `_process_payment()` ensuring no EIP-3009/Permit2 signatures are generated or executed.
+* **Docs**: Capability matrix alignment and semantic glossary updates detailing that authorization signatures are not settlement proofs.
+* **Maintained**: No payment execution behavior changed. Does not implement capture, void, refund, or reclaim lifecycles.
+
+## [1.11.0] - 2026-05-31 (Surface Preflight Read Model Client)
+* **Added**: `LnChurchClient.get_surface_preflight()` and `get_surface_preflight_async()` for safely reading historical observational memory of a surface before interacting with it.
+* **Added**: Example script (`examples/surface_preflight_read_model.py`) and unit tests to ensure strict enforcement of safety boundaries.
+* **Safety Boundaries**: No payment execution behavior was changed. The client explicitly uses isolated HTTP GET requests without engaging the `execute_detailed` HATEOAS or payment loop. Unknown surfaces safely return `known: false` without triggering HTTP 404s. The preflight result enforces `not_a_recommendation` and maintains the local runtime policy as the final authority.
+
+## [1.10.2] - 2026-05-29 (Capability Matrix & Standards Watchlist Alignment)
+* **Added**: Capability matrix documentation (`docs/07_capability_matrix.md`) mapping support boundaries and semantics.
+* **Added**: Standards watch alignment documentation (`docs/08_standards_watch_alignment.md`) clarifying upstream drift monitoring.
+* **Added**: `get_capability_matrix()` helper exposed safely in `ln_church_agent.capabilities` for tooling inspection.
+* **Maintained**: No payment execution behavior changed. L402 / x402 / MPP / SVM exact behavior remains unchanged.
+* **Maintained**: `batch-settlement` remains observe-only. AP2 / ACP / OKX APP remain inspect-only / guided-handoff surfaces. Goal Attempt telemetry remains explicit-only.
+
+## [1.10.1] - 2026-05-22 (x402 Exact EVM Hotfix)
+* **Fixed**: Prefer structured `PAYMENT-REQUIRED` over `WWW-Authenticate: x402` when both are present.
+* **Fixed**: Resolve known token-address-only x402 exact assets (Base/Polygon USDC via case-insensitive matching, Solana USDC via strict case-sensitive Base58 matching).
+* **Fixed**: Preserve raw `accepts[].amount` while exposing human-readable amount for policy checks.
+* **Fixed**: Prevent raw/human amount double conversion in EIP-3009 exact payload generation.
+* **Fixed**: Preserve structured challenge parameters such as `relayer_endpoint`.
+* **Maintained**: L402 / MPP / Payment draft priority remains unchanged.
+* **Maintained**: SVM exact raw transaction behavior remains unchanged.
+* **Maintained**: `batch-settlement` remains observe-only.
+
+## [1.10.0] - 2026-05-20 (Goal Attempt Observation & Memory)
+
+* **Added**: Introduced `submit_goal_attempt_observation()` and `submit_goal_attempt_observation_async()` for explicitly archiving goal-conditioned agent traces.
+* **Added**: Supports `goal_attempt.v1` payloads including goal declarations, attempt modes, free/paid/mixed steps, optional outcome assessments, and public-safe evidence metadata.
+* **Added**: Supports unassessed attempts. If `outcome` is omitted, the trace is preserved without forcing success/failure semantics.
+* **Added**: Lightweight Goal Attempt Memory read models allowing agents to query compact goal-scoped memory without downloading the full Monzen graph.
+* **Added**: `get_goal_attempt_summary()` / async for free summary access.
+* **Added**: `get_goal_surface_candidates()` / async for paid observed candidate surfaces.
+* **Pricing**: Goal Attempt Summary is free. Goal Surface Candidates cost 1 SAT / 0.001 USDC / 1 JPYC. Premium full graph access remains completely unchanged.
+* **Security**: Reuses strict local secret stripping while preserving public metadata such as `authorization_scheme`, `payment_performed`, and requirement fingerprints.
+* **Behavior**: Goal Attempt submission is explicit-only and does not auto-hook into `execute_detailed()`.
+* **Maintained**: Existing L402, x402, MPP, SVM, batch-settlement inspect-only, and external observation behaviors remain unchanged.
+
+## [1.9.7] - 2026-05-19 (x402 Batch Settlement Inspect-Only Support)
+* **Added**: Inspect-only awareness for x402 `batch-settlement` (a deferred settlement model separating request-time voucher authorization from batched onchain settlement).
+* **Added**: Deferred settlement metadata (`settlement_model="deferred_batch"`, `authorization_artifact="voucher"`, etc.) to `SettlementOption` and propagates them into MCP observation payloads.
+* **Security**: Explicitly blocks execution of `batch-settlement` in `_process_payment()`. The SDK safely stops without signing vouchers, depositing funds, or persisting channels.
+* **Behavior**: Classifies batch settlement as `rail="x402"` with `execution_support="observe_only"`. Voucher artifacts are not treated as settlement proofs, maintaining `payment_performed=False` and `verification_status="unverified"` in observation payloads.
+* **Maintained**: Existing L402, x402 exact, SVM exact, and MPP behaviors remain unchanged. `batch-settlement` is intentionally excluded from `SchemeType` and default `PaymentPolicy.allowed_schemes`.
+
+## [1.9.6] - 2026-05-16 (Normalized Unmapped Observation Submission)
+* **Added**: Introduced `submit_unmapped_observation()` and `submit_unmapped_observation_async()` for safely reporting unmapped or unknown payment surfaces to the observatory.
+* **Added**: Automatically normalizes `payment_scheme_unmapped`, `unsupported_challenge_shape`, and `unknown_rail` into formal, public-safe External Observation payloads.
+* **Security**: Enforces strict secret stripping (`_strip_secrets_from_evidence`) on any `extra_protocol` parameters passed to unmapped observations.
+* **Behavior**: Unmapped observations are treated strictly as discovery signals (`payment_performed=False`, `verification_status="unverified"`), not payment proofs.
+* **Behavior**: Default telemetry auto-submit remains conservative; explicit opt-in (e.g., `--include-unmapped`) is required to submit unmapped observation signals.
+* **Maintained**: The `execute_request()` unsafe HATEOAS and cross-origin navigation guardrails (`NavigationGuardrailError`) remain fully intact.
+
+## [1.9.5] - 2026-05-15 (Settlement Options Observation Layer)
+* **Added**: Introduced `SettlementOption` model for granular tracking of all payment requirements presented by an endpoint.
+* **Added**: `settlement_options` and `selected_settlement_option` fields to `InspectResult` for high-fidelity surface observation.
+* **Added**: `ln_church_observatory` metadata to `InspectResult` and CLI output to clarify local observation boundaries and opt-in submission paths.
+* **Improved**: The `inspect` tool now enumerates the full `accepts[]` array from x402 challenges without altering execution logic.
+* **Improved**: MCP observation payloads now dynamically inherit network and asset details from selected options, reducing "unknown" fallbacks.
+* **Improved**: Enhanced `missing_information` guidance for commerce surfaces (AP2/ACP/APP) when settlement rails are undeclared.
+* **Documentation**: Clarified the role of the SDK as a non-executing paid surface observer and its relationship with the LN Church Observatory.
+
+## [1.9.4] - 2026-05-14 (Payment Failure Observation Layer)
+* **Added**: Introduced `PaymentFailureRecord` for structured local recording of 402 payment attempt failures.
+* **Added**: Standardized failure taxonomy (e.g., `retry_mismatch`, `no_matching_payment_requirements`) to stabilize agent reasoning.
+* **Added**: Public-safe challenge fingerprinting and changed-field detection to identify unstable server-side requirements (e.g., dynamic `feePayer`).
+* **Added**: Payload builder for `payment_failure_observation_report.v1` for future knowledge base ingestion.
+* **Security**: Guaranteed redaction of raw secrets (macaroons, preimages, private keys) from failure artifacts.
+* **Behavior**: Non-verdict design; failures are modeled as "observed friction" rather than definitive server faults.
+
+## [1.9.3] - 2026-05-12 (MCP Registry Namespace Fix)
+* **Fixed**: Updated `mcp-name` metadata in README and `server.json` to use the `io.github.mayim-mayim` namespace, satisfying the Official MCP Registry ownership verification requirements.
+
+## [1.9.2] - 2026-05-12 (Inspect-Only MCP Entrypoint)
+* **Added**: Introduced inspect-only MCP entrypoint `ln-church-agent-mcp`.
+* **Added**: Exposes safe MCP tools for paid surface inspection, action explanation, and MCP observation payload construction.
+* **Security**: Does not require private keys or wallet configuration. Does not execute payments.
+* **Maintained**: Existing L402/x402/MPP execution behavior and standard MCP execution server (`integrations.mcp`) remain strictly unchanged.
+
+## [1.9.1] - 2026-05-11 (Guided Handoff for Agent Commerce Surfaces)
+* **Added**: Introduced "Guided Handoff" to the `inspect` CLI command. When an Agent Commerce surface (AP2, ACP, OKX APP) is detected, the SDK now provides structured guidance (`ask_site_for`, `do_not`, `required_evidence`, `missing_information`) to help AI operators safely approve or investigate the transaction.
+* **Added**: The `InspectResult` model now includes `handoff_mode`, `approval_required`, and `operator_approval_reason` to explicitly signal when human/operator intervention is necessary.
+* **Security**: Implemented strict secret-stripping for `model_dump_json()`. Raw tokens (like `shared_payment_token`, `mandate_token`, or `broker` session keys) are safely redacted and never exposed in the inspection output.
+* **Maintained**: Unchanged behavior for standard `L402`, `x402`, and `MPP` paths. Valid settlement rails will not trigger Guided Handoff unless they co-exist with a higher-order commerce surface.
+* **Details**: [v1.9.1 Release Notes](docs/release_notes/v1.9.1.md)
+
+## [1.9.0] - 2026-05-09 (AP2 / ACP Inspect-Only Commerce Surface Detection)
+* **Added**: Expanded the Agent Commerce Surface Inspector to explicitly detect Google AP2 (Agent Payments Protocol) and ACP (Agentic Commerce Protocol) metadata.
+* **Added**: Introduced `surface_type`, `surfaces_detected`, `settlement_rails_detected`, `detection_confidence`, `detection_reason`, and `unsupported_reason` to `InspectResult` for granular classification.
+* **Changed**: Strictly decoupled Commerce Surfaces (AP2, ACP, OKX APP) from executable Settlement Rails (L402, x402, MPP). AP2/ACP metadata are treated strictly as authorization/commerce mandates, not settlement proofs.
+* **Behavior**: Default `recommended_action` for AP2 and ACP is `observe_only`. The SDK intentionally **does not** execute these mandates.
+* **Security**: Introduced a `stop_safely` guardrail if an AP2/ACP surface co-exists with a malformed or unsupported settlement hint.
+* **Maintained**: 100% backward compatibility for existing L402, x402, and MPP execution behaviors.
+
+## [1.8.5] - 2026-05-08 (Sandbox Evidence Corpus Readiness)
+* **Added**: `SandboxCorpusCandidate` model to lightly transform sandbox execution evidence into a local candidate format.
+* **Added**: `build_sandbox_corpus_candidate` helper to perform eligibility checks (e.g. `non_sandbox_scope`, `candidate_pending_client_confirmation`).
+* **Added**: `client.get_last_sandbox_corpus_candidate()` and `client.build_sandbox_corpus_candidate_from_last_evidence()`.
+* **Added**: Preserves rail metadata such as `network`, `asset`, `payment_method`, `authorization_scheme`, and `draft_shape` in corpus candidates.
+* **Maintained**: Guaranteed no automatic submissions to `ExternalObserve` telemetry API to prevent Sandbox isolation leakage. Final corpus acceptance remains server-side.
+
+## [1.8.4] - 2026-05-08 (Sponsored Access & Sandbox Evidence Alignment)
+* **Added**: `SponsoredAccessEvidence` model to safely capture JWS grant consumption metadata (`grant_jti`, `issuer`, `scope`) with strict architectural constraints (`settlement_rail: "none"`).
+* **Added**: `SandboxEvidence` model to capture isolated testbed telemetry from `sandbox_evidence_ref.v1` and `sandbox_evidence_report.v1`.
+* **Added**: Optional fields `sponsored_access` and `sandbox` to `PaymentEvidenceRecord` for automatic EvidenceRepository exports.
+* **Added**: Safe builder helpers and `sha256_redacted` to ensure raw `interop_token` and `grant_token` secrets are permanently excluded from JSON models and exports.
+* **Added**: New public APIs including `get_last_sponsored_access_evidence()`, `get_last_sandbox_evidence()`, and `get_sandbox_evidence_logs()`.
+* **Maintained**: Guaranteed no automatic submissions to `ExternalObserve` telemetry API to prevent Sandbox isolation leakage.
+
+## [1.8.3] - 2026-05-08 (Sponsored Access Diagnostics)
+* **Added**: Introduced `GrantDiagnostics` to locally pre-evaluate JWS grant tokens.
+* **Added**: Added `diagnose_grant()` and `explain_grant()` to `LnChurchClient` for AI-friendly payload generation.
+* **Added**: Added `ln-church-agent grant inspect` command to the CLI.
+* **Clarified**: Formally decoupled Grants from settlement rails (`settlement_rail: "none"`).
+* **Maintained**: Guaranteed graceful fallback behavior for invalid grants without interrupting the core L402/x402 loop.
+
+## [1.8.2] - 2026-05-07 (Response Adapter Decompression Hotfix)
+* **Fixed**: Resolved `httpx.DecodingError` during `inspect` by safely stripping hop-by-hop headers (`Content-Encoding`, `Transfer-Encoding`, `Content-Length`) from the response adapter.
+* **Improved**: Ensures `inspect` fails gracefully with a structured `response_decoding_error` diagnostic instead of completely crashing the pipeline when upstream metadata is severely malformed.
+
+## [1.8.1] - 2026-05-07 (Inspector Robustness & OpenClaw Stability)
+* **Fixed**: Resolved execution environment dependencies for the `inspect` CLI. It now supports stable worker execution via direct Python API calls or `sys.executable`.
+* **Improved**: Hardened the x402 challenge parser to capture non-standard shapes, such as Alchemy-style challenges that place `accepts` or `resource` fields directly within the JSON body.
+* **Added**: Advanced classification logic for the `Payment` scheme, automatically mapping it to `MPP` or `x402` rails based on the provided `payment_method`.
+* **Improved**: Enhanced `InspectResult` diagnostics by returning detailed `error_stage`, `failure_reason`, and `diagnostic_class` fields to facilitate automated system registration decisions.
+* **Fixed**: Introduced `_safe_float` to prevent crashes when the `amount` field contains non-numeric strings or objects, ensuring they are safely handled as `0.0`.
+
+## [1.8.0] - 2026-05-06 (Agent Commerce Surface Inspector & APP Detection)
+* **Added**: Introduced the **Agent Commerce Surface Inspector** capability to the CLI `inspect` tool.
+* **Added**: Safe detection for OKX Agent Payments Protocol (APP) metadata without payment execution.
+* **Added**: Extensible commerce classification fields: `commerce_protocol`, `commerce_intent`, `commerce_transport`, `authorization_artifact`, `settlement_rail`, `settlement_method`, and `broker_required`.
+* **Added**: Automated settlement rail normalization (e.g., mapping `scheme: exact` to `rail: x402`).
+* **Behavior**: Challenge co-existence support. Can now detect both a Commerce layer and its underlying Settlement rail simultaneously (e.g., `rails_detected: ["APP", "x402"]`).
+* **Architecture**: Aligned internal models with future Agent Commerce standards (Google AP2, ACP, UCP) as a buyer-side observation layer.
+* **Maintained**: 100% backward compatibility for existing `L402`, `x402`, and `MPP` paths.
+* **Details**: [v1.8.0 Release Notes](docs/release_notes/v1.8.0.md)
+
+## [1.7.3] - 2026-05-03 (External Observation Client & x402 Exact Diagnostics)
+* **Added**: `run_x402_evm_exact_sandbox_diagnostic()` and `run_x402_svm_exact_sandbox_diagnostic()` to test post-settlement validation of V2 exact envelopes.
+* **Added**: `submit_external_observation()` and `get_external_observations()` for protocol-level telemetry, with strict local stripping of raw secrets.
+* **Changed**: `parse_challenge_from_response` now strictly parses Hybrid V1+V2 challenge shapes, resolving `token_address`, `decimals`, and `accepts[].asset` accurately.
+* **Changed**: `inspect` CLI command now recognizes `scheme: "exact"` and recommends `observe_only` to prevent unintended settlement loops.
+* **Architecture**: Formalized that the current LN Church x402 exact sandbox acts as a *post-settlement validator*. Unbroadcasted payloads are intentionally rejected (Expected Rejection). True V2 exact settlement (where the facilitator broadcasts) is a future phase.
+* **Details**: [v1.7.3 Release Notes](docs/release_notes/v1.7.3.md)
+
+## [1.7.2] - 2026-04-30 (First Success UX - Inspect CLI)
+* **Added**: Introduced the `inspect` CLI command to analyze HTTP 402 challenge structures (Rail, Intent, Shape) and provide recommended actions without executing payment.
+* **Added**: Support for the short command alias `lnc-agent`.
+* **Added**: `InspectResult` model for CLI analysis with first-class `--json` output support.
+* **Added**: Stability tests for the CLI using mocked networks in `tests/test_cli_inspect.py`.
+* **Details**: [v1.7.2 Release Notes](docs/release_notes/v1.7.2.md)
+
+## [1.7.1] - 2026-04-30 (Agent-Side Synthetic Corpus Replay Runner)
+* **Added**: `LnChurchClient.run_corpus_replay()` and async counterpart for dry-run validation of Server Synthetic Corpus Replays.
+* **Added**: `CorpusReplayResult` model to encapsulate the comparison between `expected_action` and `observed_action`.
+* **Details**: [v1.7.1 Release Notes](docs/release_notes/v1.7.1.md)
+
+## [1.7.0] - 2026-04-29 (Official x402 v2 SVM Exact Path)
+* **Added**: Native buyer-side runtime support for official x402 v2 SVM exact payments (`scheme: "exact"`, `network: "solana:<genesisHash>"`).
+* **Added**: Native SVM Exact Transaction Builder to construct x402 v2 compatible VersionedTransaction payloads natively.
+* **Added**: Optional `[svm]` and `[all]` extra dependency flags for `solana` and `solders` libraries.
+* **Added**: `allowed_networks` constraint to `PaymentPolicy` for strict CAIP-2 gating.
+* **Changed**: Legacy `lnc-solana-transfer` remains safely isolated as an LN Church compatibility path.
+* **Details**: [v1.7.0 Release Notes](docs/release_notes/v1.7.0.md)
+
+## [1.6.5] - 2026-04-29 (Challenge Parser Boundary Refactor)
+* **Changed (Internal)**: Extracted HTTP 402 challenge parsing and Payment/MPP challenge shape classification into a dedicated internal module (`ln_church_agent/challenges.py`).
+* **Compatibility**: Preserved `Payment402Client` compatibility wrappers. No public API changes were introduced, and all payment execution, settlement logic, and v1.6.4 telemetry behaviors remain strictly unchanged.
+* **Details**: [v1.6.5 Release Notes](docs/release_notes/v1.6.5.md)
+
+## [1.6.4] - 2026-04-29 (Payment Draft Telemetry & Interop Observation)
+* **Added**: Payment HTTP Authentication draft-aware parsing. The SDK safely parses `Payment` scheme challenges and decodes Base64URL JSON requests to extract underlying invoices.
+* **Added**: Advanced Interop Matrix telemetry for MPP. Challenges are now classified by `draft_shape` (e.g., `payment-auth-draft`, `legacy-mpp-flat`), `payment_method`, and `payment_intent`.
+* **Changed**: MPP session intent is successfully observed and classified, but execution is safely halted (`mpp_session_not_supported_yet`) to prevent unverified runtime flows while still capturing the telemetry.
+
+## [1.6.3] - 2026-04-27 (MPP Charge Sandbox Harness Integration)
+* **Added**: Introduced `run_mpp_charge_sandbox_harness()` and `run_mpp_charge_sandbox_harness_async()` to `LnChurchClient` to seamlessly validate the IETF draft Machine Payments Protocol (MPP) flows.
+* **Changed**: Upgraded Interop Report telemetry for *both* MPP and L402 harnesses to include dynamic protocol metadata (`rail`, `payment_intent`, `authorization_scheme`, `payment_receipt_present`), enhancing observability on the public Interop Matrix.
+* **Fixed**: Replaced hardcoded authorization schemes in the Sandbox Harness with dynamic extraction from the `SettlementReceipt` to accurately absorb protocol fluctuations.
+* **Added**: Introduced `test_v1_6_3_mpp_sandbox.py` to ensure regression protection for dynamic telemetry extraction.
+* **Docs**: Updated Sandbox documentation and README to reflect the availability of the parallel MPP Charge endpoint.
+
+## [1.6.2] - 2026-04-26 (x402 V2 Compatibility Improvements & Payload Normalization)
+* **Added**: Support for transparent protocol extensions. The SDK captures server-provided extension metadata (like discovery configurations) and echoes it back in the settlement payload to facilitate upstream indexers.
+* **Added**: Integrated standard `"exact"` scheme support into the core loop, formatting `PAYMENT-SIGNATURE` headers properly for gasless V2 settlements.
+* **Changed**: Upgraded 402 challenge parsing to handle the x402 V2 `accepts` array structure, extracting matching network options based on the agent's expected `chainId`.
+* **Fixed (Critical)**: Reconstructed the V2 `PAYMENT-SIGNATURE` envelope to match expected V2 hierarchy (`x402Version: 2`, `accepted`, `resource`, `payload`, `extensions`), resolving rejection loops.
+* **Fixed (Critical)**: Resolved a fatal RPC failure in EVM transfers by replacing an invalid `eth_price` call with `eth_gasPrice` and properly padding raw transactions.
+* **Fixed**: Implemented practical heuristics for unit normalization (Raw-to-Human), converting minimal-unit integers (e.g., Wei) to human-readable decimals to prevent false-positive blocks by local policies.
+* **Fixed**: Enhanced treasury address (`payTo`) extraction from nested V2 structures and implemented safe fallbacks to expected logical symbols (USDC/JPYC) for raw contract addresses.
+* **Details**: [v1.6.2 Release Notes](docs/release_notes/v1.6.2.md)
+
+## [1.6.1] - 2026-04-24 (EVM Signature & CDP Compatibility Patch)
+* **Added**: Enhanced `LnChurchClient` convenience methods (`draw_omikuji`, `submit_confession`, etc.) with `**kwargs` support for direct, seamless payload parameter injection (e.g., `chainId`).
+* **Fixed**: Resolved `invalid_payload` failures in EVM gasless settlement (`lnc-evm-relay`) by extracting `r` and `s` signatures directly from the 65-byte hex string, guaranteeing strict 64-character zero-padding.
+* **Fixed**: Added the missing `to` (treasury address) field to the relayer payload to ensure complete EIP-712 data transmission and full compatibility with strict verifiers like Coinbase CDP.
+* **Fixed**: Explicitly included `chainId` in the relayer payload to enable seamless cross-chain gasless settlements (e.g., Base and Polygon) and prevent incorrect network defaulting.
+* **Details**: [v1.6.1 Release Notes](docs/release_notes/v1.6.1.md)
+
+## [1.6.0] - 2026-04-22 (Internal Access Selection & Refactoring)
+* **Changed (Internal)**: Refactored the internal access selection loop for `LnChurchClient` by introducing strict Selector and Builder separation (`_ExecutionAccessPlan`, `_FundingPolicy`, etc.).
+* **Compatibility**: Guaranteed 100% backward compatibility with the 1.5.x public API, concrete vocabulary (`GRANT_CREDIT`, `grant`, `faucet`), and wire-level protocol.
+* **Details**: [v1.6.0 Release Notes](docs/release_notes/v1.6.0.md)
+
+## [1.5.12] - 2026-04-21 (Sponsored Access Override)
+* **Added**: Introduced `set_grant_token()` to the `LnChurchClient` to hold a sponsor-issued JWS grant.
+* **Added**: Implemented local JWT pre-evaluation (`has_valid_scoped_grant`) to verify expiration, audience, and route scopes client-side without cryptographic overhead.
+* **Added**: Expanded `AssetType` with `GRANT_CREDIT` and `SchemeType` with `grant`.
+* **Added**: `examples/use_grant_omikuji.py` to demonstrate the End-to-End sponsored access flow.
+* **Changed**: Upgraded `draw_omikuji` and `draw_omikuji_async` to natively prioritize valid Grant tokens as a `paymentOverride`, gracefully falling back to legacy Faucet or standard 402 challenges if the token is invalid or expired.
+* **docs(1.5.12)**: synchronize architecture roadmap with current STANDARDS_WATCHLIST and standards-tracking policy (no code changes)
+
+## [1.5.11] - 2026-04-17 (Interop Matrix Separation & Live Diagnostics)
+* **Added**: `comparison_class` and `test_mode` to Interop reports. Explicitly separates "Intentional Mismatches" (validation tests) from production errors in the ledger and UI.
+* **Fixed**: Resolved an attribution bug where `executor_mode` (Native/Delegated) defaulted to Native during payment failures. Mode is now determined pre-execution.
+* **Added**: Introduced diagnostic fields (`suspected_failure_origin`, `upstream_host_excerpt`) to `ExternalProtocolRunResult`. Enables heuristic identification of infrastructure errors like Cloudflare 520.
+* **Added**: Implemented `run_external_protocol_verification` (Sync/Async) helpers for benchmarking live, unmanaged L402 endpoints.
+* **Changed**: Extended LNBits settlement polling buffer from 1s to 5s to account for typical live Lightning Network routing latencies.
+
+## [1.5.10] - 2026-04-16 (The Advisor & Final Judge Architecture)
+* **Changed**: Refactored `RemoteTrustEvaluator` and `RemoteOutcomeMatcher` to act as final judges that synthesize remote advice with local agent policies, rather than blindly delegating decisions.
+* **Added**: The LN Church backend now acts as an "Evidence-Rich Advisor", returning `recommendation`, `checks`, and `evidence_bundle` instead of centralized verdicts.
+* **Changed**: Renamed `fallback_mode="unknown"` to `allow_on_error` in `RemoteTrustEvaluator` to accurately reflect its fail-open behavior.
+* **Added**: Local policy overrides (e.g., `allowed_hosts`) and custom local fallback matchers now strictly supersede remote backend recommendations.
+
+## [1.5.9] - 2026-04-15 (Evidence-Backed Session Budget Persistence)
+* **Added**: Introduced `session_spend_delta_usd` to `PaymentEvidenceRecord` to capture immutable settlement budget events.
+* **Added**: Expanded `EvidenceRepository` with `import_session_evidence` hooks to enable session budget recovery across agent crashes or restarts.
+* **Added**: Built-in cryptographic deduplication using `receipt_id` to prevent double-counting of session budgets during HATEOAS recovery loops.
+* **Changed**: Centralized the USD exchange rate logic into an internal `_estimate_usd_value` helper to ensure consistency across policy enforcement and evidence recording.
+* **Changed**: Optimized the core execution loop (`execute_detailed` / `async`) to perform a lightweight, one-shot session budget restore prior to standard HATEOAS navigation.
+
+## [1.5.8] - 2026-04-15 (Navigation Hint Normalization & Observability)
+* **Added**: Introduced "Navigation Hint Normalization" to absorb HATEOAS vocabulary fluctuations (e.g., `next`, `action`, `retry_action`) and Header-based hints (`Location`, `Link`) into a canonical `NextAction` model.
+* **Added**: Enhanced the observability layer by adding `navigation_source` to `PaymentEvidenceRecord`, allowing agents to audit the decision origin of HATEOAS recovery paths.
+* **Changed**: Fortified Cross-Origin guardrails to use strict **netloc** (host:port) matching, preventing malicious redirections to unauthorized ports on trusted domains.
+* **Changed**: Implemented "Header Hardening" to automatically isolate and strip sensitive credentials (e.g., `Authorization`) from server-suggested headers during autonomous navigation.
+* **Fixed**: Eliminated redundant `asyncio.sleep(1)` from the asynchronous execution path to ensure wire-level performance parity between Sync and Async runtimes.
+* **Fixed**: Synchronized the public API surface by exporting `ChallengeSource` and aligning `ParsedChallenge` with mandatory schema fields, resolving latent `NameError` in downstream validation tests.
+
+## [1.5.7] - 2026-04-14 (Documentation Alignment & Protocol Fluctuation Absorption)
+* **Fixed (Docs)**: Resolved a documentation misalignment from v1.5.6 regarding the `Payment` and `MPP` header prefixes.
+* **Changed**: Formally documented the SDK's dynamic protocol absorption capability. The client transparently supports both IETF Draft (`Payment`) and ecosystem (`MPP`) standards by dynamically constructing the `Authorization` header based on the server's `WWW-Authenticate` challenge, adhering perfectly to the Cold Spec governance.
+* **Details**: [v1.5.7 Release Notes](docs/release_notes/v1.5.7.md)
+
+## [1.5.6] - 2026-04-14 (Wire-Level Protocol Purity & Interface Sync)
+* **Fixed**: Resolved a critical parsing paradox where `MPP` headers in `WWW-Authenticate` were ignored, ensuring proper dual-stack routing.
+* **Fixed**: Restored protocol purity for Lightning payments (L402/MPP) by preventing `PAYMENT-SIGNATURE` and JSON body pollution, strictly using the `Authorization` header.
+* **Fixed**: Synchronized the `EVMSigner` interface in `protocols.py` with the canonical v1.5.2 naming conventions to prevent `AttributeError` for custom wallet adapters.
+* **Changed**: ~~Standardized the MPP authorization header output to use the `MPP` prefix instead of the legacy `Payment` prefix.~~ **(※RETRACTED IN v1.5.7: The SDK dynamically supports both `Payment` and `MPP` prefixes based on server requirements, rather than hardcoding. See v1.5.7 notes.)**
+* **Details**: [v1.5.6 Release Notes](docs/release_notes/v1.5.6.md)
+
+## [1.5.5] - 2026-04-13 (Dual-Stack Resilience & Initialization Fix)
+* **Fixed**: Reordered 402 challenge parsing to prioritize Lightning (L402) over x402, resolving the "Dual-Stack Paradox" where L402 invoices were ignored.
+* **Fixed**: Normalized `ValueError` bubbling and corrected constructor argument propagation in `LnChurchClient`.
+* **Fixed**: Improved the legacy challenge parser to fetch missing parameters (like destination) from the response body.
+* **Details**: [v1.5.5 Release Notes](docs/release_notes/v1.5.5.md)
+
+## [1.5.4] - 2026-04-13 (Wire-Level Standard Compliance)
+* **Changed**: Transitioned x402 payment headers from legacy string-concatenation to standard Base64URL-encoded JSON objects (Payment Payload / Payment Required / Settlement Response).
+* **Fixed**: Improved `_parse_challenge` and `_extract_receipt` to prioritize Base64-encoded JSON parsing while maintaining backward compatibility with legacy string-based headers.
+* **Added**: Native `_b64url_decode` and `_b64url_encode` helpers in the core client for robust payload handling.
+* **Details**: [v1.5.4 Release Notes](docs/release_notes/v1.5.4.md)
+
+## [1.5.3] - 2026-04-13 (x402 Standard Stabilization)
+* **Fixed**: Resolved critical execution blockers (missing imports, `TypeError` in signature, `NameError` in policy enforcement) introduced during the 1.5.2 x402 Foundation alignment.
+* **Fixed**: Restored legacy header parsers (`WWW-Authenticate`, `x-402-payment-required`) to ensure backward compatibility while migrating to standard x402.
+* **Added**: Comprehensive strict-mode tests for the full `PAYMENT-REQUIRED` to `PAYMENT-RESPONSE` autonomous negotiation roundtrip.
+* **Details**: [v1.5.3 Release Notes](docs/release_notes/v1.5.3.md)
+
+## [1.5.2] - 2026-04-12 (x402 Foundation Alignment)
+* **Changed**: Achieved full compliance with x402 Foundation (Linux Foundation) standards and CAIP-2 network identifiers.
+* **Changed**: Updated `LnChurchClient` defaults to `L402` / `SATS`, prioritizing Lightning-native settlement.
+* **Changed**: Normalized custom routing identifiers to the `lnc-` prefix (`lnc-evm-transfer`, `lnc-solana-transfer`, `lnc-evm-relay`).
+* **Added**: Internal normalization layer (`_normalize_scheme`) for legacy scheme alias resolution to maintain backward compatibility.
+* **Added**: Regression tests for standard compliance and convenience defaults.
+* **Deprecated**: Legacy scheme Enum members (`x402-direct`, `x402-solana`).
+* **Fixed**: Removed legacy vocabulary from specification artifacts (`openapi.yaml`, `agent-api.json`) and documentation.
+* **Details**: [v1.5.2 Release Notes](docs/release_notes/v1.5.2.md)
+
+## [1.5.1] - Experimental Evidence Export/Import Layer
+* **Added**: `EvidenceRepository` base class with `export_evidence` and `import_evidence` hooks (sync/async).
+* **Added**: `PaymentEvidenceRecord` to safely encapsulate the lifecycle of a 402 interaction (intentionally excluding secrets like preimages).
+* **Changed**: Core execution engine automatically imports past evidence into `context.past_evidence` and exports records upon completion or failure.
+* **Details**: [v1.5.1 Release Notes](docs/release_notes/v1.5.1.md)
+* **Fixed**: Normalized `ValueError` bubbling during client initialization. Invalid private keys now consistently return a unified, predictable error message regardless of the underlying cryptographic adapter, squashing a latent initialization bug.
+
+## [1.5.0] - Source-Agnostic Trust & Provider-Agnostic Outcome
+* **Added**: `TrustEvidence` model to abstract trust evaluation inputs (URL, metadata, agent hints).
+* **Changed**: `OutcomeMatcher` can now accept `SettlementReceipt` to perform cross-verification between the payment proof and the host's response.
+* **Changed**: `ExecutionContext` now supports `hints` for passing top-down agent knowledge into hooks.
+* **Compatibility**: Evaluators and Matchers written for v1.4 remain 100% backward compatible via dynamic signature inspection.
+* **Details**: [v1.5.0 Release Notes](docs/release_notes/v1.5.0.md)
+
+## [1.4.0] - Trust & Outcome Layer (Decide & Verify)
+* **Added**: `TrustEvaluator` hooks to evaluate counterparty risk before payment.
+* **Added**: `OutcomeMatcher` hooks to semantically verify expected outcomes after execution.
+* **Added**: `ExecutionContext` for lightweight session and intent tracking.
+* **Details**: [v1.4.0 Release Notes](docs/release_notes/v1.4.0.md)
+
+## [1.3.1] - Async Performance & UX Patch
+* **Fixed**: Reused `httpx.AsyncClient` to prevent socket exhaustion in high-frequency runtimes.
+* **Fixed**: Replaced silent identity fallback with explicit `ValueError` on bad private keys.
+* **Details**: [v1.3.1 Release Notes](docs/release_notes/v1.3.1.md)
+
+## [1.3.0] - Safety & Stability Overhaul
+* **Fixed**: `PaymentPolicy` type safety and precise session budget accounting.
+* **Added**: Backward compatibility wrapper for `execute_paid_action`.
+* **Details**: [v1.3.0 Release Notes](docs/release_notes/v1.3.0.md)
+
+## [1.2.x] - Economic Guardrails & Risk Verification
+* Introduced `PaymentPolicy` limits, `SettlementReceipt` generation, and Keyless Agent execution via `NWCAdapter`. Added MCP tools for Counterparty Risk Verification.
+* **Details**: [v1.2.5](docs/release_notes/v1.2.5.md), [v1.2.4](docs/release_notes/v1.2.4.md), [v1.2.3](docs/release_notes/v1.2.3.md)
+
+## [1.1.0] - Dynamic EVM Auto-Routing
+* Enhanced `x402` schemes with Dynamic Multi-Chain Auto-Routing and Solana Standards alignment.
+
+## [1.0.0] - Initial Stable Release
+* Introduced the autonomous `Probe → Pay → Execute` loop across `L402`/`MPP` (Lightning), `x402` (Polygon), and `x402-solana` (Solana).

mayim_agent-2.0.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 ln-church
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.