Agents

Added

• Added a dev-only Webhook Test Sink with local capture storage, signature verification, CP inspection, a one-click Send test webhook action, and scripted smoke/E2E validation helpers for local webhook development.
• Added env-aware webhook target and signing-secret fields to Settings so runtime webhook transport can be configured from the CP using Craft-native env/alias inputs.

Changed

• Realigned the CP information architecture and canonical routes around Status, Approvals, Accounts, and Settings, with the visible paths now using /status, /approvals, and /accounts.
• Hardened first-run operator UX so healthy fresh installs bias toward Ready to Connect, treat sync-state as optional until configured, and keep confidence gaps visible without making the whole page pessimistic.
• Reworked the webhook test sink into a dedicated Status card with capture-state handling, payload drill-down, and clearer dev-only/runtime-target copy.
• Standardized rectangular CP card surfaces to a 3px radius and tightened Status card composition, diagnostics-bundle placement, and proof/action affordances.
• Removed Discovery Docs from the core plugin surface, including CP UI, routes, generated discovery files, diagnostics references, and docs coverage.

Fixed

• Fixed the webhook payload dialog so long payloads scroll inside the modal instead of expanding the entire overlay.

Changed

• Reworked the top-level CP information architecture to Status, Approvals, Accounts, Discovery Docs, and Settings, removing local sidebars from Status, Accounts, and Discovery Docs and promoting Discovery Docs to its own top-level surface.
• Added a first-run Ready to Connect bootstrap state for fresh installs so healthy but inactive environments no longer open on a pessimistic Unproven verdict.
• Renamed the CP-facing Control surface to Approvals while preserving the underlying governed-write routes and compatibility redirects.
• Reframed account-level webhook subscription copy as event-interest routing so operators can more clearly understand how external workers are selected and woken through the shared webhook destination.
• Added a Monthly Report Agent managed-account template and tightened Discovery Docs/Status presentation to match the current CP direction.

Changed

• Merged dashboard security posture fully into Readiness, so operators now work from one combined state card, one action-mapping table, and one shared proof-card grid.
• Replaced the separate security summary/proof surfaces with merged proof-card detail dialogs for Traffic / Access, Delivery / Webhooks, Integration / Capacity, Credentials / Policy, and Confidence / Observability.
• Removed the standalone Security dashboard tab while preserving legacy route and anchor compatibility inside the merged Readiness surface.
• Tightened the merged readiness card styling to match the current Figma direction for the top signal header, summary strip, proof-card borders, and embedded detail actions.
• Removed the readiness-page security technical JSON section now that security posture is represented through the merged proof-card and detail-dialog model.

Changed

• Reworked the Dashboard Readiness view into an operator state card with summary strips, structured proof panels, and a filtered action-mapping table that only appears when signals need follow-up.
• Reworked the Dashboard Security view to use the same state-card and action-mapping model while preserving dead-letter queue replay operations below the summary surface.
• Added focused deep-link support on Accounts cards so dashboard remediation links can open and highlight the most relevant machine account context.
• Added stable section anchors for Dashboard and Settings surfaces and updated dashboard regression checks to match the current card-based CP architecture.

Fixed

• Fixed the Security dashboard dead-letter queue summary so an empty queue no longer throws a Twig runtime error when rendering the latest-update field.

Changed

• Refined the Control CP tables to share a consistent Waiting for Decision-derived header/body treatment across approvals, follow-up, activity, and rules views.
• Added collapsed-by-default disclosure toggles for Approved, Applied / Completed, Runs That Need Follow-up, Activity Log, and inline Proposed changes details.
• Tightened Control CP spacing and card-strip behavior for a more consistent Craft-native operator experience across desktop and mobile.

Changed

• Hardened machine-write auth: query-token transport remains read-only even when enabled, and write routes now require header auth plus Content-Type: application/json.
• Bound sync-state checkpoint writes to the authenticated credential context so dedicated credentials can no longer overwrite another integration's checkpoint state.
• Persisted approval assurance mode and downgrade reason on each request (dual_control, single_approval, single_operator_degraded) so later operator-count changes do not rewrite historical approval strength.
• Surfaced approval assurance details in the Control CP and control-approval flash messaging for clearer operator auditability.

Fixed

• Fixed requester/approver separation so self-approval is blocked whenever a request was evaluated under non-degraded assurance.
• Fixed managed credential generation to fail closed when random_bytes() is unavailable instead of falling back to predictable entropy.

Added

• Added guarded runtime incident feed endpoint GET /agents/v1/incidents with incidents:read scope and query filters (severity, limit) for strict-redacted reliability incident snapshots.
• Added incident snapshot coverage to runtime contracts (/capabilities, /openapi.json, /schema) and reliability regression checks.

Changed

• Updated release/docs parity for the new incidents scope and endpoint across README and operator runbooks.
• Updated Accounts scope selection defaults to include incidents:read for managed credential setup.
• Hid local /.tmp workspace artifacts from release surfaces by adding /.tmp/ to .gitignore.

Fixed

• Fixed GET /agents/v1/openapi.json response maps so OpenAPI responses are emitted as status-code objects (not arrays), restoring validator compatibility for GPT Actions and other OpenAPI tooling.
• Fixed OpenAPI POST operation contracts to include a minimal JSON requestBody.content schema where bodies are required, preventing requestBody validation errors in strict Action importers.
• Added an absolute API server URL (https://<host>/agents/v1) to the OpenAPI servers list so GPT Actions can resolve a valid server URL without manual schema edits.

Changed

• Unified Control CP and governed write APIs behind one gate: PLUGIN_AGENTS_WRITES_EXPERIMENTAL (removed separate CP override behavior).

Changed

• Renamed governed draft-write scope to entries:write:draft for clarity, and kept entries:write as a deprecated compatibility alias.
• Simplified governed-approval operator flow: final approval now executes immediately, and approved/apply/completed states are clearer in the Control CP.
• Relaxed dual-approval requirement when only one active CP user exists, so single-operator installs can still run governed flows.

Added

• Added governed write action support for entry.updateDraft through POST /agents/v1/control/actions/execute, including native draft creation/update execution in ControlPlaneService.
• Added action payload contract metadata for entry.updateDraft to OpenAPI/schema descriptors (x-action-payloads / xActionPayloads).
• Added experimental entries:write scope for governed entry-draft updates and exposed it in Accounts scope selection (effective only when PLUGIN_AGENTS_WRITES_EXPERIMENTAL=true).
• Added reference automation fixture + docs for governed entry draft updates:
  • docs/reference-automations/fixtures/entry-update-draft-execute.json
  • docs/reference-automations.md
  • docs/canonical-first-agent-jobs.md
• Added template + starter-pack discoverability for governed-entry-draft-update across template/starter services and docs.
• Added per-account write indicator icons in CP cards: locked icon when write actions require human approval, unlocked icon when write actions are allowed without approval.
• Added per-account “Always require human approval” control for write-capable accounts in CP create/edit flows.

Changed

• Extended QA regression scripts to assert governed entry-draft write contracts and starter-pack/reference automation coverage.
• Added a CP Settings runtime toggle (enableWritesExperimental) for governed writing/control API surfaces, with env/config lock handling.
• Removed legacy refund/return env-flag aliases; governed write surfaces now respond only to PLUGIN_AGENTS_WRITES_EXPERIMENTAL and PLUGIN_AGENTS_WRITES_CP_EXPERIMENTAL.
• Restricted CP visibility/effectiveness of per-account human-approval control to write-capable accounts only.

Fixed

• Fixed CP per-account human-approval lightswitch persistence and edit-mode hydration so off/on state is saved and displayed correctly.
• Fixed write indicator rendering and status copy on account cards for write-capable credentials.

Changed

• Migrated sync-state naming across docs/contracts/QA from legacy /consumers/* wording to canonical /sync-state/* endpoints.
• Refined CP dashboard and accounts UX with a unified metric-strip card style, improved card filtering states, and updated discovery-doc card interactions.
• Updated Reliability Threshold settings fields to use Craft env-var-aware inputs in CP.

Fixed

• Fixed reliability threshold parsing so numeric and env-var-backed values are persisted and evaluated consistently in runtime signals.
• Fixed reliability QA coverage for env-var-driven threshold settings.

Added

• Added per-agent Owner input in Agents CP create/edit form, with create-mode default prefilled from current CP user email.

Changed

• Persisted credential owner metadata in managed credential storage and fed it into lifecycle ownership posture (with .env metadata-map fallback retained).
• Added migration m260306_100000_add_credential_owner_column and regression coverage for owner-field flows.
• Simplified Agents CP cards by removing lifecycle inline warning strips and the lifecycle risk table block.

Fixed

• Fixed token-authenticated machine POST compatibility by disabling CSRF enforcement on API controller endpoints (including /agents/v1/consumers/checkpoint).

Changed

• Hidden lifecycle warning surfaces from the Agents CP view by removing the Lifecycle Governance summary block and per-agent warning strips/risk labels from cards.
• Kept lifecycle governance backend/API/CLI data paths intact so warning UI can be reintroduced without schema or service rollback.
• Updated lifecycle governance QA assertions to reflect the new CP visibility contract.

Added

• Added canonical template catalog service with API endpoint GET /agents/v1/templates (templates:read) and CLI command craft agents/template-catalog.
• Added schema/OpenAPI-linked reference automation docs and JSON fixtures for the three canonical first jobs.
• Added dedicated regression check (scripts/qa/reference-automations-regression-check.sh) and integrated it into the release gate.
• Added starter-pack catalog service with API endpoint GET /agents/v1/starter-packs (templates:read) and CLI command craft agents/starter-packs for copy/paste runtime snippets (curl, javascript, python).
• Added integration starter-pack docs at docs/integration-starter-packs.md.
• Added reliability threshold evaluation service with read-only triage summaries embedded in GET /agents/v1/metrics.
• Added CLI reliability snapshot check (craft agents/reliability-check) with strict mode support for CI/operator gates.
• Added dedicated reliability regression check (scripts/qa/reliability-pack-regression-check.sh) and integrated it into the release gate.
• Added lifecycle governance service with API endpoint GET /agents/v1/lifecycle (lifecycle:read) and CLI command craft agents/lifecycle-report.
• Added lifecycle governance operator docs (docs/agent-lifecycle-governance.md) and VitePress troubleshooting page for ownership/risk posture workflows.
• Added dedicated lifecycle governance regression check (scripts/qa/lifecycle-governance-regression-check.sh) and integrated it into the release gate.

Changed

• Extended capability/openapi/schema contracts to advertise and describe template catalog usage for integrators.
• Enriched diagnostics bundle output with reliability summary/signal snapshots for faster incident triage.
• Enriched diagnostics bundle output with lifecycle status/summary snapshots for ownership and stale-agent triage.
• Updated Dashboard Readiness tab with “Needs Attention Now” triage signals and threshold-driven runbook guidance.
• Updated observability runbook thresholds and response playbooks for reliability signals.
• Updated Agents CP view with lifecycle governance summary cards and per-agent risk factor visibility.
• Hid CP Control tab/routes/permissions by default behind an internal CP-only flag (PLUGIN_AGENTS_WRITES_CP_EXPERIMENTAL) while keeping control-plane API/data internals unchanged.

Changed

• Reordered Control Panel subnavigation to place Agents directly below Dashboard for faster operator access.

Added

• Added CP runtime setting enableCredentialUsageIndicator to toggle live per-agent usage activity indicators on the Agents cards.
• Added managed-agent pause state persistence (pausedAt) with migration m260305_110000_add_credential_pause_column.
• Added pause/resume lifecycle actions for managed agents in CP and runtime credential filtering.

Changed

• Reworked the CP Agents view from API-key table workflows to card-based agent management with inline create/edit flows.
• Renamed CP navigation and permission copy from API-key terminology to agent terminology.
• Extended managed credential usage tracking to classify read/write operations for activity-state UI feedback.

Fixed

• Fixed pause/resume action reliability across upgraded installs by handling missing pause-column scenarios safely.
• Fixed live usage indicator behavior to respect settings while still allowing explicit debug simulation via query params.

Added

• Added environment profile resolver with optional PLUGIN_AGENTS_ENV_PROFILE (local|test|staging|production) and inferred profile fallback.
• Added profile-based runtime defaults for auth/rate-limit/webhook posture when explicit PLUGIN_AGENTS_* values are unset.
• Added runtime profile metadata across health/readiness/capabilities/schema/diagnostics outputs (environmentProfile, environmentProfileSource, profileDefaultsApplied, effectivePolicyVersion).
• Added read-only CP Environment Profile posture visibility in Security views.

Added

• Added inventory-aware product snapshots: GET /agents/v1/products now includes hasUnlimitedStock and totalStock per item.
• Added low-stock filtering on GET /agents/v1/products via lowStock and lowStockThreshold query parameters (full-sync mode).
• Added inventory fields to variant list payloads so GET /agents/v1/variants now exposes stock, hasUnlimitedStock, and isAvailable.

Added

• Added read APIs for additional Craft and Commerce resources: users (flag-gated), assets, categories, tags, global sets, addresses (flag-gated), content blocks, variants, subscriptions, transfers, and donations.
• Expanded GET /agents/v1/changes coverage to include newly exposed resources for broader incremental-sync parity.
• Added canonical agent handbook link exposure in discovery outputs (/llms.txt, /llms-full.txt).

Added

• Added one-click diagnostics bundle foundation across API (GET /agents/v1/diagnostics/bundle), CP download flow, and CLI (craft agents/diagnostics-bundle).
• Added diagnostics:read scope and contract metadata updates across capabilities/OpenAPI/schema/readme docs.

Fixed

• Fixed release metadata alignment by publishing a fresh immutable patch version after the previous v0.6.1 tag pointed at a pre-bump commit.
• Fixed plugin-version fallback constants in API/readiness telemetry to match the current release.

Added

• Added release-gate protections to detect stale runtime version fallbacks before publish.

Fixed

• Fixed adoption instrumentation runtime fatal by switching to the existing security posture API used across CP/runtime.
• Fixed machine-client POST compatibility by disabling CSRF enforcement for token-authenticated API endpoints.
• Fixed dual-approval race handling by adding optimistic concurrency guards/retries in approval decision flow.
• Fixed stale plugin-version fallback constants in API/readiness outputs.

Added

• Added guarded observability export endpoint GET /agents/v1/metrics (metrics:read) with runtime counters for auth failures, scope denials, rate-limit denials, request volume, and 5xx responses.
• Added CP Readiness telemetry snapshot cards sourced from observability metrics, plus threshold-based runbook/alert guidance for incident triage.

• Improved CP operations UX with clearer section grouping, full-width separators, and state color coding across overview/dashboard/control/credentials views.
• Added API contract hardening via deterministic query validation (400 INVALID_REQUEST with details) for malformed fields/filter, enum, numeric, and identifier query paths.
• Added adoption instrumentation endpoint GET /agents/v1/adoption/metrics with adoption:read scope for funnel, time-to-first-success, and weekly managed-credential usage signals.
• Added canonical QA gates for API/scope/docs parity, deterministic validation regression, control/consumer surface regression, and migration safety checks; integrated into scripts/qa/release-gate.sh.
• Added operator-facing adoption docs: canonical first agent jobs and a copy/paste 30-minute quickstart flow.

• Added webhook dead-letter queue persistence with guarded API replay endpoints and Control Panel replay controls.
• Added per-credential webhook subscription targeting (resource/action filters) to reduce firehose delivery.
• Added consumer lag tracking surfaces (/agents/v1/consumers/checkpoint, /agents/v1/consumers/lag) and dashboard visibility.
• Added credential expiry policies (TTL + reminder windows), CP warnings, and runtime exclusion of expired managed keys.
• Added credential CIDR allowlists with runtime API auth enforcement and CP management UI.
• Added policy simulator dry-run flow (/agents/v1/control/policy-simulate) plus CP simulation tooling.
• Added two-person approval support for high-risk actions with staged approval progress tracking.
• Added SLA escalation and auto-expiry behavior for pending approvals, including surfaced SLA state in CP/API payloads.
• Added list endpoint projection/filter support (fields, filter) on key read surfaces to reduce payload size.
• Added versioned machine-readable schema catalog endpoint (/agents/v1/schema) for safer client generation.

• Added optional extended discovery export GET /llms-full.txt with capabilities/OpenAPI/CP surface alignment.
• Added CP-editable custom body settings for llms.txt and commerce.txt, including config lock-state awareness.
• Added Settings actions to reset custom llms.txt and commerce.txt bodies back to generated defaults.
• Extended discovery cache/prewarm/status flows to cover llms-full.txt and invalidate on settings save.
• Updated Dashboard discovery tab copy/labels to Discovery Docs for clearer IA.

• Hardened discovery contract by adding root discovery aliases (/capabilities, /openapi.json) that map to the guarded API descriptors.
• Added authenticated introspection endpoint GET /agents/v1/auth/whoami with scope visibility, auth method details, and rate-limit snapshot.
• Added CLI validation commands for operators and CI:
  • craft agents/auth-check
  • craft agents/discovery-check
  • craft agents/readiness-check
  • craft agents/smoke
• Updated capabilities/OpenAPI/README contract metadata to include new auth/discovery surfaces and CLI checks.
• Updated llms.txt discovery output to annotate auth requirements/scopes and include canonical alias pointers.
• Dashboard settings now respect config/agents.php overrides for discovery toggles (enableLlmsTxt, enableCommerceTxt) and display lock-state guidance in CP.

• Improved Dashboard tab readability by using the active tab label as the page heading.
• Removed repeated service-state panels from Readiness, Discovery, and Security tabs to reduce duplicate status noise.
• Refined Discovery tab document panels: default-at-a-glance status now focuses on URL and Last Modified, additional metadata is collapsed under Details, and preview code blocks use a subtle bordered style.

• Fixed CP navigation state so the Agents section/subnav remains active across Dashboard, Settings, API Keys, and Control routes.
• Fixed plugin settings entry-point behavior: opening Agents from admin/settings/plugins now redirects to admin/agents/dashboard/overview.
• Added an Agents discovery caches option to Craft’s Clear Caches utility (agents-discovery) to clear cached llms.txt and commerce.txt documents.
• Added canonical CP redirects for admin/agents and admin/agents/dashboard to admin/agents/dashboard/overview.

• Polished CP IA by consolidating Overview/Readiness/Discovery/Security into a Dashboard with top tabs, while preserving legacy deep links via redirects.
• Renamed and simplified the experimental approvals area to Control with clearer queue-first copy (Now, decisions, follow-up runs, activity) and agent-first fallback messaging.
• Improved API Keys UX with preset examples, native Craft scope selection, one-time key copy/download helpers, and a revoke+rotate shortcut action.
• Improved CP readability by default-collapsing technical JSON blocks and tightening labels/messages across settings and credential actions.
• Updated readiness diagnostics to treat CP and site web contexts as valid request context for the web-request readiness check.

• Hid governed-write/control surfaces behind PLUGIN_AGENTS_WRITES_EXPERIMENTAL (default off): CP tab/routes, API routes, capabilities/OpenAPI discoverability, and related scope catalog entries are now gated.
• Added agent-first approval request mode: CP request form is disabled by default and can be re-enabled via settings.
• Added API scope split for approvals: control:approvals:request and control:approvals:decide (legacy control:approvals:write remains supported).
• Added required approval-request provenance metadata on API (metadata.source, metadata.agentId, metadata.traceId).

• Reworked the Control Plane CP interface into a queue-first operator flow (Now, Act, Configure, Audit).
• Added guided approval/execution forms with optional advanced JSON overrides for payload and metadata.
• Added policy-aware execute guardrails in CP (disabled-policy blocking, approval-required validation, action-type match checks).
• Improved control action flash messaging for idempotent replay, approval decisions, and blocked/failed execution outcomes.

• Added control-plane foundation with governed policies, approvals, idempotent action execution ledger, and immutable audit events.
• Added new guarded control API endpoints under /agents/v1/control/* with explicit read/write scopes for policies, approvals, executions, and audit access.
• Added Control Plane Control Panel tab with policy upsert, approval queue decisions, execution controls, and control-plane snapshot visibility.
• Added plugin migration for control-plane persistence tables: policies, approvals, executions, and audit log.

• Hardened Commerce availability checks in discovery/readiness surfaces to use project config state instead of forcing Commerce plugin bootstrap.
• Reduced risk of early getCurrentStore() fatals during plugin startup in environments where Commerce store/site mappings are incomplete.

• Added Control Panel credential lifecycle foundation: managed credential create/edit scopes/rotate/revoke/delete flows and one-time token reveal UX.
• Added managed credential persistence with runtime auth integration and last-used metadata tracking (lastUsedAt, lastUsedIp, auth method).
• Added permission-granular credential actions (view, manage, rotate, revoke, delete) and updated CP/docs posture to reflect hybrid env + managed credential support.

• Promoted incremental sync capabilities to the v0.2.0 minor baseline.
• Finalized deterministic cursor/updatedSince continuation behavior on /products, /orders, /entries, and /changes.
• Finalized queued webhook delivery with X-Agents-Webhook-Signature (HMAC-SHA256) and bounded retry semantics.
• Completed OpenAPI/capabilities alignment and integrated incremental/webhook regression harnesses into release gating.

• Hardened incremental request validation on /products to return deterministic 400 INVALID_REQUEST for malformed cursor/updatedSince inputs.
• Tightened credential parsing in SecurityPolicyService to accept only credential-object shapes and ignore malformed scalar entries.
• Added explicit Cache-Control: no-store, private headers for guarded JSON and API error responses.
• Normalized order change snapshots in /changes to use updatedAt consistently.
• Expanded OpenAPI route response metadata to include guarded error outcomes (401/403/429/503) across protected endpoints.
• Added incremental and webhook regression harnesses and integrated them into the release gate workflow.
• Added release validation evidence and handoff checklist updates for maintainers.

• Added request correlation IDs on API responses via X-Request-Id.
• Standardized error response schema with stable error codes and per-response requestId/status.
• Added capabilities/OpenAPI error taxonomy metadata for integration clients.
• Added INCREMENTAL_SYNC_CONTRACT.md defining cursor/updatedSince, ordering, replay, and tombstone semantics for v0.2.0.
• Added incremental sync filters to /products, /orders, and /entries with cursor precedence, deterministic ordering, and snapshot-window continuation metadata.
• Added GET /agents/v1/changes unified feed with normalized created|updated|deleted items, deterministic checkpoint continuation, and tombstones from soft-deleted records.
• Added optional webhook delivery for product|order|entry change events with queued retries and X-Agents-Webhook-Signature HMAC verification headers.
• Added CP cockpit IA v1 with 4 deep-linkable tabs: overview, readiness, discovery, and security (legacy agents/dashboard + agents/health aliases retained).
• Added shared SecurityPolicyService for effective auth/rate-limit/redaction/webhook posture across API, CP security view, and plugin startup warnings.
• Added discovery operator controls in CP (prewarm all|llms|commerce, clear cache) with read-only discovery metadata/previews.
• Fixed CP template resolution using Craft plugin-handle template conventions (agents/* in CP mode).

• Hardened API auth defaults with explicit production fail-closed behavior.
• Disabled query-token auth by default and clarified token transport metadata.
• Added scope-aware authorization for sensitive order and non-live entry access.
• Added discovery text generation for /llms.txt and /commerce.txt with cache + ETag/Last-Modified behavior.
• Added cache invalidation hooks and CLI prewarm command: craft agents/discovery-prewarm.
• Improved rate-limiting strategy with pre-auth throttling and atomic counter fallback.

• Initial public release of klick/agents.
• Added read-only HTTP API endpoints for products, orders, entries, sections, readiness, and health.
• Added discoverability endpoints: /agents/v1/capabilities and /agents/v1/openapi.json.
• Added read-only CLI commands under craft agents/*.
• Added control panel section for plugin dashboard and health views.