Release notes from AgentAction
Articles about Release notes from AgentAction
AgentAction 0.50.0-rc.1
Added Automatic Workato synthetic-refund policy path: eligible USD refunds up to $100 execute with scope-bound single-use JIT authorization and signed outcome verification, without a reviewer or approval request.
AgentAction 0.49.0-rc.1
Added Workato governed-refund reference integration with a private Connector SDK connector, separate recipe and reviewer roles, durable approval/status handling, and a synthetic-provider execution adapter.
AgentAction 0.48.0-rc.1
Added Newly saved social scanners execute the frozen, server-validated v2 workflow contract: bounded operations, typed input connections, AI instructions and parameters, retrieval/poll/deadline limits, report bounds and stable delivery identity.
AgentAction 0.47.0-rc.1
Added Read-only social research workflow inspection in My agents and Runs: typed step connections, AI instructions/model, provider bindings, limits, failure behavior and evidence associated with stable step IDs.
AgentAction 0.46.0-rc.1
Added Research landing page for Can We Trust 'Done'?, including its abstract, manuscript PDF, experimental figure, author attribution, reproducibility folder, and an invitation to submit community review comments.
AgentAction 0.45.0-rc.1
Added Optional second daily local research time, shown in saved cards and the explicit schedule review. The scheduler chooses the next local slot across timezone and daylight-saving changes. Indep
AgentAction 0.44.3-rc.1
Fixed Recognize Apify’s exact shortened tool name for the long X research Actor while retaining its full execution identity. Freeze validation rules without descriptive schema annotations for the
AgentAction 0.44.2-rc.1
Fixed Explicitly select Apify’s run-status and dataset-reading helpers alongside the two research Actors. Initial discovery now requests the complete workflow tool set; missing tools are named in the error.
AgentAction 0.44.1-rc.1
Fixed Research setup now reviews existing Apify credential reuse directly in the page, avoiding native confirmation dialogs that can stall embedded browsers. Changing the selected account clears the review.
AgentAction 0.44.0-rc.1
Added A reviewed X and Reddit daily research workflow in My agents. Saved company briefs produce editable search phrases; explicit section saves preserve the relevance scope, recipient, local scan time and provider limits.
Release v0.43.0-rc.1 · dinpd/AgentAction
AgentAction 0.43.0-rc.1
AgentAction 0.42.0-rc.1
Added Definable preparation skills with reusable workspace versions, local customization and explicit version adoption. The company research default reads a public website and drafts product, audience, problems, search phrases, brand queries and exclusions with source references.
AgentAction 0.41.2-rc.1
Fixed Read streamed MCP responses incrementally and stop at the matching JSON-RPC response instead of waiting for the connection to close. Handle fragmented events, UTF-8, progress notifications and batched SSE responses while retaining the response-size limit.
AgentAction 0.41.1-rc.1
Fixed Compare freshly discovered tool definitions in the same credential-scrubbed representation used by saved catalogs. Continue to block genuine definition changes, validate against the live input schema, and identify changed definition sections without exposing provider values.
AgentAction 0.41.0-rc.1
Added Public official-registry MCP directory and stable server profiles at mcpcheck.agentaction.dev/servers, linked from the website and checker. Search, review setup declarations, inspect existing public observations, and suggest corrections without a workspace login.
AgentAction 0.40.2-rc.1
Fixed Put current runs and pending approvals above agent cards and group completed history by agent. Retain each agent’s latest result alongside recent history so frequent health checks cannot hide it. Show the proposed action before technical evidence.
AgentAction 0.40.1-rc.1
Fixed Show Save changes, Saving…, Saved and Retry save directly on every guided draft save button. Disable already-saved and pending buttons, expose pending saves to assistive technology, and keep all section buttons synchronized when edits arrive during a save.
AgentAction 0.40.0-rc.1
Added Explicit Save draft controls at the bottom of every setup section, saving all current draft edits with visible unsaved, saving, saved and retryable failure states. No autosave or execution is implied. Warn before leaving unsaved edits; concurrent saves cannot replace newer edits.
AgentAction 0.39.1-rc.1
Fixed Explain selected tools as needing a supervised trial instead of showing an unexplained Unknown warning for optional field checks. Keep technical findings expandable and distinguish unavailable tools and field conflicts from unverified behavior.
AgentAction 0.39.0-rc.1
Added Guide draft setup through job details, sources, guardrails and checks, and a supervised trial. Show the next missing step prominently, retain editable sections, and tuck alternative entry points and source options away until needed.
AgentAction 0.38.1-rc.1
Fixed Label Apify Actor credentials as a required API token, link to Apify Console, and explain where to find the token and how to enter it. Anonymous endpoint pre-checks remain available without credentials.
AgentAction 0.38.0-rc.1
Added Resolve Apify Actor listings to documented hosted MCP endpoints, including existing catalog snapshots. Selecting a server fills its Actor-scoped endpoint, starts an anonymous pre-check, and explains the API-token requirement with a setup source link.
AgentAction 0.37.2-rc.1
Fixed Recognize combined generic processing labels such as text analysis and summarization, so they do not become redundant MCP setup requirements alongside external capabilities. Preserve specialized/provider terms, source retrieval and concrete tool bindings.
AgentAction 0.37.1-rc.1
Fixed Preserve unanswered scope questions when selecting a provider and returning from connection setup. Keep generic text analysis, reasoning and report writing inside the agent instead of adding
AgentAction 0.37.0-rc.1
Added Generate editable pass thresholds, measurement procedures and evidence sources with agent drafts. Freeze them into trial contracts and require the AI assessor to report observed measurements; missing evidence remains inconclusive.
AgentAction 0.36.1-rc.1
Fixed Assign generated outcome-check IDs in the application. Missing, duplicated or reserved model-authored IDs no longer reject otherwise valid guardrails and evaluation criteria; labels and criteria remain intact and validated.
AgentAction 0.36.0-rc.1
Added Generate job-specific guardrails and outcome evaluation criteria together with provider-neutral capability requirements. Connected inventory no longer shapes the draft objective or preselects tools.
AgentAction 0.35.3-rc.1
Fixed Make the selected MCP server name and endpoint bold in both connection panels, with primary text contrast and wrapping for narrow screens. Compatibility and migration Backward-compatible presentation-only patch prerelease; no behavior, permission, schema or migration changes. Install Python artifacts from the GitHub v0.35.3-rc.1 prerelease. Hosted services deploy from the merge commit; no registry publication is implied.
AgentAction 0.35.2-rc.1
Fixed Restore live Notion OAuth discovery, verified with 45 account tools. Support large MCP tool schemas without truncating their validation rules. Input and output schemas are bounded at 128 KiB and the complete catalog at 512 KiB; the 80-tool count and response limits remain enforced.
AgentAction 0.35.1-rc.1
Fixed Accept OAuth token responses that omit expiration, as allowed by Notion's client contract, using a one-hour local lease. Cap longer advertised lifetimes at one year; malformed explicit lifetimes remain rejected. Expired local leases refresh or require owner reconnection.
AgentAction 0.35.0-rc.1
Added Owner-managed OAuth accounts for shared workspace MCP agents, with PKCE, authenticated callbacks, encrypted credentials, refresh rotation and disconnect/revocation. Operators can use shared accounts through existing action approvals and schedules; grants remain workspace-owned after the connecting member leaves.
AgentAction 0.34.2-rc.1
Changed Make the existing MCP readiness checker available at https://mcpcheck.agentaction.dev, with website navigation and a homepage entry point for MCP developers. Replace Recipes in the primary
AgentAction 0.34.1-rc.1
Fixed Restore public MCP checker form initialization in production. Disable Wrangler name-preservation helpers that cannot be referenced from the serialized browser script. Run readiness acceptanc
AgentAction 0.34.0-rc.1
Added Public MCP Readiness Check with anonymous HTTPS discovery, protocol selection, source-linked schema findings, usability guidance, metadata heuristics, JSON export and explicit opt-in publication.
AgentAction 0.33.0-rc.1
Added Optional Smithery and Glama directory enrichment for MCP discovery before account connection. Search locally cached tool names, descriptions and input/result schemas; show potential matching tools, fields, source links, retrieval dates and unknown/partial/stale metadata.
AgentAction 0.32.0-rc.1
Added Discover MCP servers within each required draft capability, including unconnected registry servers and relevant connected tools. Show match reasons, provider details and connection state, with editable searches and a connect-your-own action for every capability.
AgentAction 0.31.0-rc.1
Added Show MCP capabilities and limits in the observability console, separating public pre-check summaries from connected-account snapshots. Retain bounded result schemas, annotations, resource listings and discovery provenance; older snapshots remain unknown until refreshed.
AgentAction 0.30.0-rc.1
Added Capture paginated MCP tools, resources, and resource templates with bounded discovery, explicit incomplete/unknown states, catalog hashes, and server/protocol provenance. Discovery does not execute tools or read resources.
AgentAction 0.29.0-rc.2
Fixed Generate profiler ideas solely from the selected function/area and optional improvement context. Existing MCP tools no longer bias suggestions toward unrelated work or block discovery due to endpoint availability.
AgentAction 0.29.0-rc.1
Added Replace generic Create example prompts with optional Help me choose an agent. Choose or type a function or area, optionally describe an improvement, and receive up to three tailored agent ideas with benefits, job descriptions and capability needs.
AgentAction 0.28.0-rc.2
Fixed Simplify Create to one job-description form with clickable example prompts. Browse catalog and saved examples through an optional disclosure, show Continue an agent only when drafts exist, and remove the competing server-based creation section.
AgentAction 0.28.0-rc.1
Added Create an agent before choosing MCP servers. Describe a job or use an agent template, save an unbound workspace draft, and return to finish its tools later. Create is the first lifecycle step; MCP servers live in workspace navigation.
AgentAction 0.27.0-rc.1
Added Describe a job once in Create to generate an editable agent draft from the selected connection's discovered tools. Prefill its name, reusable objective, instructions, expected result and standard measurable checks, while preserving the original description as instance inputs.
AgentAction 0.26.0-rc.1
Added Define measurable checks inline in Create and retain them with workspace recipe revisions. Check successful tool use, absence of a tool call, and typed assertions on the latest tool call's MCP structuredContent. Completion, successful execution, selected-tool scope, recorded approval and the four-call budget are checked for every bound run.
AgentAction 0.25.0-rc.1
Added Author agent recipes directly in Create, starting from scratch, a catalog recipe, an AI suggestion or a saved workspace recipe. Edit the goal, input guidance, allowed tools, instructions, boundaries and success criteria together, then create a supervised draft with its own job inputs.
AgentAction 0.24.0-rc.1
Added Browse Activity by agent in compact, expandable groups, with current status, last/next execution and findings. Pending approvals and current problems sort first. Switch to a chronological list, filter by agent or execution status, and page through retained history ten executions at a time.
AgentAction 0.23.0-rc.2
Fixed Show recurring recipe instances alongside supervised agents in My agents, with latest execution, result and next schedule. Include recurring checks in Run history and expose both kinds of run in Monitor, newest first. Keep pending approvals visible beyond the recent-history limit.
AgentAction 0.23.0-rc.1
Added Reusable recurring-agent handlers with owner-approved intervals, durable baselines, run evidence, restart recovery, findings and acknowledge/resolve/reopen transitions. An observed outage is a completed monitoring run with a finding; stale or incomplete checks remain unknown.
AgentAction 0.22.0-rc.3
Added Consolidate recipe adoption into Create. Public recipe pages open the same pinned catalog recipe in the console; review requirements, connect a missing MCP server, choose its connection, and configure a draft without generating AI suggestions. Preserve recipe selection across sign-in reload, workspace changes and workspace setup while clearing private draft inputs when switching workspaces.
AgentAction 0.22.0-rc.2
Fixed Separate Workspace settings from the numbered agent lifecycle, with a dedicated utility destination on both console and builder. Overview remains the progress home and guides new users into workspace setup. Existing setup and invitation URLs keep working, including the selected workspace.
AgentAction 0.22.0-rc.1
Added Restructure AgentAction into Overview, Connect, Create, Run, Monitor and Improve. The home screen shows workspace progress and a next action; the builder separates provider setup, agent creation, and supervised execution into focused screens. Monitor brings together activity, finalized jobs and execution quality; Improve holds evals.
AgentAction 0.21.0-rc.1
Added Separate Browse servers and MCP connections views with preserved search filters and pagination. Selecting a server opens focused setup with automatic credential-free pre-checks; valid manual endpoints are checked after a short pause. Findings remain visible alongside approval and connection actions, with local errors and recovery controls.
AgentAction 0.20.0-rc.1
Added Run a credential-free endpoint pre-check before approving or connecting an MCP provider. Registry cards and manual entry lead to prominent, saved workspace findings with the exact endpoint, observation time, OAuth discovery, provider metadata, advertised/challenged scopes, public tool descriptions, HTTP evidence and explicit uncertainty.
AgentAction 0.19.0-rc.1
Added Filter My Agents registry results by declared API keys, Authorization headers, other secret inputs or Not specified, alongside capability and text search. Result cards show the declared authentication labels; filters remain consistent across pagination.
AgentAction 0.18.0-rc.4
Fixed My Agents now shows the signed-in account, selected workspace role, Log out and Workspace setup in a persistent account section. Roles come from workspace membership; switching accounts does not assign a new role.
AgentAction 0.18.0-rc.3
Fixed Show actionable connection blockers immediately beside Connect: missing endpoint, owner approval or view-only access. Disabled My Agents buttons now show an unavailable cursor. A busy cursor and accessible busy state appear only while a request is running, so missing endpoint approval no longer looks like a stuck connection.
AgentAction 0.18.0-rc.2
Fixed Cached MCP Registry cards now display current workspace-owner approval guidance immediately after deployment, without waiting for a catalog refresh. Provider metadata and endpoint authorization are unchanged. Install Python artifacts from the GitHub v0.18.0-rc.2 prerelease. This continues the v0.18.0 authorization release candidate; no migration or npm publication is required.
Release v0.18.0-rc.1 · dinpd/AgentAction
AgentAction 0.18.0-rc.1
AgentAction 0.17.0-rc.1
Discover MCP servers during onboarding Search the official MCP Registry by server name, description or advertised capability, with category filters, paged results and provider setup links in My Agents.
AgentAction 0.16.0-rc.1
MCP agent instances Release candidate for the new runtime storage migration and AI execution surface. Add an authenticated My Agents builder: connect an approved HTTPS MCP endpoint with optional b