Skip to content

CHANGELOG.md

Latest commit

 

History

History
172 lines (100 loc) · 21 KB

CHANGELOG.md

File metadata and controls

172 lines (100 loc) · 21 KB
Raw

Changelog

All notable changes to @salesforce/mulesoft-vibes-skills are documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[1.8.4] - 2026-08-10

Fixed

  • build-agent-broker-project and translate-agent-broker-old-to-new-project — Removed the incorrect claim that connection ids (context.connections.<id>) and broker ids (keys under brokers:) are restricted by the V2 schema to ^[a-z0-9_]+$ (lowercase/digits/non-trailing-underscore only) and that camelCase or kebab-case "fails lint." The schema does not enforce a snake_case-only format on these keys — any valid YAML identifier validates. The skills now frame snake_case as a readability convention and preserve the one genuine invariant: an .agent target (a2a://, mcp://, llm://, brokers://) must match its corresponding agent-network.yaml key exactly. Affects canonical-example.md item 0, gotchas.md naming conventions, and the converter's connection/broker translation notes, template comments, and example annotations.

Changed

  • build-agent-broker-project — Updated the Step 9 deploy/gateway guidance to the current Anypoint CLI Agent Fabric plugin (v1.2.10) gateway model: single-gateway mode (-g/--gateway, default agent-network-gw) is the recommended path, with separate ingress/egress (-i/--ingress-gw + -e/--egress-gw) as the alternate mode. Replaces the stale separate-gateway-only defaults (agent-network-ingress-gw / agent-network-egress-gw).

[1.8.3] - 2026-07-13

Fixed

  • manage-api-version — Step 9's rescaffold (mvn clean package -DskipTests) required a manual approval click in the VS Code Vibes/Agentforce UI even when the exact command was allowlisted, because the model was emitting it as cd <projectDir> && mvn clean package -DskipTests 2>&1. The 2>&1 stream redirection is flagged as an unsafe operator by the extension's command-approval layer (A4dSafeCommandsController.containsUnsafeOperators()), which forces manual approval unconditionally, before any allowlist pattern is even consulted. Step 9 now explicitly instructs the agent never to append 2>&1 or any other redirection/pipe to the mvn invocation — the tool call already captures full stdout/stderr, so the redirection was redundant as well as harmful. Also made explicit that this rescaffold step must run immediately with no user confirmation prompt, matching the real Anypoint Studio behavior where rescaffolding fires automatically on a pom.xml change or a Project Properties → API Specs tab edit and is never a user-facing decision.

Changed

  • Added cd ... and an exact-match mvn clean package -DskipTests line to the local a4d_safe_commands allowlist (developer machine config, not part of this repo) so the two sub-commands produced by Step 9 — split independently by the extension's parseMultiCommand() — both auto-approve once the 2>&1 redirection is removed.

[1.8.2] - 2026-07-13

Fixed

  • manage-api-version — Step 7c ("present available versions and handle selection") told the agent to display the version list and prompt as plain narrated chat text, then separately invoke the AskUserQuestion tool for the actual interactive selection — producing two renderings of the same question in the VS Code Vibes/Agentforce UI (one plain-text, one interactive). Step 7c is rewritten so the version list and prompt are presented only via a single AskUserQuestion tool call per API, with the version list as the tool's options; the skill no longer narrates the same content as chat text before or after the tool call.

[1.8.1] - 2026-07-13

Fixed

  • manage-api-version — fixed a false-negative bug in version discovery (Steps 5, 6, 7b). The skill previously relied solely on anypoint-cli-v4 exchange asset describe <groupId>/<artifactId>/<currentVersion> and its otherVersions field to enumerate an asset's available versions. That field is anchored to the queried version and has been observed to return empty when the queried version is not the latest published version on Exchange — exactly the case CHECK ALL/CHECK SPECIFIC/CHANGE hit every time, since they query from whatever version is currently pinned in pom.xml. This caused the skill to report "no updates available" or "only one version published" even when newer versions genuinely existed. Version discovery now uses anypoint-cli-v4 exchange asset list <artifactId> --output json filtered to the target groupId/assetId as the primary lookup — an asset-identity-scoped query with no version-anchoring failure mode, matching the pattern already used by build-mule-integration/scripts/search_templates.sh and manage-global-configurations/scripts/get_latest_connector.sh. The old describe-based lookup is retained only as a degraded fallback when list itself fails, with an explicit caveat surfaced to the user that the resulting list may be incomplete.

Changed

  • manage-api-version — added a bundled scripts/fetch_versions.sh helper and rewired the Version Discovery procedure (Steps 5, 6, 7b) to call it once per operation instead of looping and invoking anypoint-cli-v4 once per API dependency. Each direct CLI invocation pays a full Node cold start plus a network round trip; the new script fires the Exchange lookup for every target API in parallel and returns a single JSON array, cutting wall-clock time roughly N-fold on projects with multiple API dependencies. Same pattern already used by build-mule-integration/scripts/search_templates.sh and manage-global-configurations/scripts/get_latest_connector.sh.

[1.8.0] - 2026-06-30

Added

  • manage-api-version — new skill for reading, checking, and updating API spec dependency versions in a Mule project. Reads <{artifactId}.version> properties from pom.xml, queries Anypoint Exchange for available versions via anypoint-cli-v4 exchange asset describe, and applies version changes followed by an automatic mvn clean package -DskipTests rescaffold. Supports four paths: display all versions, display specific versions, check all APIs for newer versions, and a full interactive change flow with semver-sorted version selection, pom.xml backup/restore on failure, and batched multi-API updates in a single Maven pass.

[1.7.0] - 2026-07-16

Added

  • author-governance-ruleset — new skill that authors valid Anypoint API Governance rulesets (Validation Profile 1.0 YAML) using the anypoint-cli-v4 governance:ruleset CLI for model discovery, validation, and simplification. Force-installs the latest governance plugin, resolves domain language to canonical target classes, discovers classes/properties/constraints per domain, then writes, validates (validate-authoring), and simplifies the ruleset before presenting it. Covers OpenAPI, RAML, AsyncAPI, MCP servers, Anypoint API instances, and API projects; enforces the single-specKind rule and never guesses class/property names or constraint compatibility.

1.6.1 — 2026-07-15

Changed

  • build-agent-broker-project and translate-agent-broker-old-to-new-project — enforce snake_case for connection IDs and broker IDs in skill code examples per the V2 schema (^[a-z0-9_]+$). Prior examples emitted camelCase / kebab-case which the ACB linter rejects. Also loosens the agent_name rule to "optional, conventionally kebab-case" since the docs mark it optional and the field has no strict format. Verified against the authoritative docs at mulesoft/docs-code-builder@latest/agent-network/2.0/modules/ROOT.

[1.6.0] - 2026-07-02

Added

  • generate-bat-tests — new skill that generates a runnable BAT (Blackbox API Testing) BDD suite — DataWeave .dwl files plus a bat.yaml manifest — from a Mule app's source (OpenAPI contract + Mule flow XML), organized by quality dimensions (Accuracy / Robustness / Security / Coverage) and validated by running it against the live HTTP endpoint. BAT is black-box and out-of-process (it hits the deployed API, never imports flows) and has no XSD, so the workflow's gate is "the suite parses AND passes against the running app" rather than a build-time compile. Two-phase workflow with a hard approval gate: Phase 1 reads the source, anchors an endpoint/raise-error allowlist, and drafts a dimension-tagged test inventory for the user to approve; Phase 2 scaffolds the suite, writes the .dwl files, statically validates the BAT DSL, and runs ./run-bat.sh against the endpoint. Bundles four scripts (validate_prerequisites.sh, extract_endpoints.sh, scaffold_suite.sh, validate_bat_suite.sh) and reference material (BAT authoring rules, the quality-dimension taxonomy, and canonical .dwl + bat.yaml examples). Can also extend an existing hand-written BAT baseline as a strict superset. Complements MUnit generation (build-time, in-process, XML) by covering deployed-endpoint functional testing.

[1.5.0] - 2026-06-25

Changed

  • build-mule-integration — synced from the agent-evaluation lab to skill v12.2.0 internal tag. Reorders Step 1b options so "I want to generate from scratch without a template." is the first option (default-safe). Adds explicit Steps E5 / L5 continue-or-stop checkpoints at the end of the Exchange and Local sub-flows so the agent always asks the user whether to proceed to connector discovery + flow generation or stop after template setup. Local template format relaxed from .jar-only to .jar or .zip (the underlying dx:mule:project:create --template-file CLI accepts both).
  • build-mule-integration — adds an absolute-path / "no relative ../scripts/..." rule for invoking bundled scripts, plus a "why scripts instead of inline bash" rationale (loss of resolved GAVs across Bash tool calls), based on production-run failure analysis.
  • build-mule-integration — Phase 2 step ranges renumbered (Steps 8–17 from 8–18); flow XML generation cross-references corrected (Steps 10/12 instead of 11/13); pre-mvn validation script reference added to the troubleshooting cheatsheet.

[1.4.0] - 2026-06-23

Added

  • generate-connectivity-knowledge — new 14-step skill that produces connectivity knowledge for a SaaS API when no dedicated Mule connector exists. Researches the API from user-defined use cases and documentation, generates an OpenAPI 3.0 spec, validates every operation against the live service with auto-fix, and writes a self-contained connectivity-schema/<apiName>/ folder (api-reference.md + <apiName>.yaml + config.properties). Output feeds the HTTP-fallback branch of build-mule-integration so HTTP-Connector flows inherit the same auth, pagination, and entity awareness a dedicated connector would carry.

[1.3.0] - 2026-06-20

Added

  • mulesoft-agent-broker-builder — new skill that drives an end-to-end Agent Network V2 (GA, A2A v1.0) build experience: 6-phase guided requirements → asset registration → Agent Script authoring → instruction refinement → topology review, plus optional publish and deploy. CLI-first via the Anypoint CLI Agent Fabric plugin (agent-network project create/build/publish/deploy/setup-gateways) with the MuleSoft MCP server as fallback and a graceful no-tool degradation path. Step 0 invokes agent-network project create to produce a starter project with the correct groupId/organizationId, then the skill edits files in place. Bundles a canonical IT Help Investigation example (sourced from the working stgx-it-investigation-GA-ver reference) and a gotchas reference covering A2A v1.0 vs v0.3 (a2a_v03) backward-compatible card branches, GA echo update events (a2a:status_update_event / a2a:artifact_update_event with TASK_STATE_* enum), compile-error rules for action invocation (A2A bare reference vs with message = in executors; MCP inputs:/with/slot-fill rules), connection authentication (required on kind: llm), policies as {inbound, outbound} object, subagent-vs-orchestrator decision, CR-18 least-privilege binding, RULE-ASSET-MODE (inline vs Exchange registration), and the full CLI / MCP capability matrix with env-var auth.
  • mulesoft-agent-broker-v1-to-v2-converter — new skill that converts an Agent Network V1 project (schemaVersion: 1.0.0) into a V2 project (agentNetwork: 2.0.0). Each V1 broker becomes a V2 broker backed by an Agent Script .agent file with one orchestrator node — preserves the user's prompt verbatim, does not split into routers/executors/generators. V1 agents land in the metadata.interfaces.a2a_v03 (back-compat) branch; the broker emits A2A v1.0. Bundles a canonical V1 input (customer-onboarding-v1) and matching V2 output, plus a v2-template.agent skeleton. For richer multi-node graphs, the skill points users at mulesoft-agent-broker-builder as the natural next step.

[1.2.1] - 2026-06-12

Changed

  • build-mule-integration — absorbed create-project-template as a conditional sub-file (references/template-project-creation.md). Template-based project creation (Exchange search, local .jar) is now a branch within Step 1b, loaded only when the user wants a template. search_templates.sh moved to build-mule-integration/scripts/.
  • build-mule-integration — Step 8 dx mule project create example now passes --skip-environment for scratch projects to avoid an unnecessary environment-resolution API call.

Removed

  • create-project-template — removed as standalone skill. Its workflow lives in build-mule-integration/references/template-project-creation.md.

[1.1.1] - 2026-06-09

Changed

  • build-mule-integration — broader connector-search guidance so private (UUID-groupId) connectors published to a customer's Exchange tenant surface alongside public ones, and the prose now tells the agent to escalate via AskUserQuestion when both a public and a private connector match the same system family. Step 3's "Common search terms" table uses broader system names (salesforce, database, http, netsuite, servicenow, jms, slack) so private assets whose assetId does not share tokens with the public connector still surface.
  • build-mule-integration — Step 16 gains a pre-mvn static validator (scripts/validate_before_build.sh) that checks the connector error-type allowlist (Cluster D), namespace ↔ pom.xml dependency parity (Cluster A2-A5), and canonical XSD URL shape — fast line-numbered diagnostics instead of a 30 s+ Maven failure.
  • build-mule-integration — scripts/describe_connector.sh now caches per-connector and per-operation errorTypes to tmp/connector-errors/, which the new validator reads.
  • build-mule-integration — scripts/get_latest_connector.sh ranking/scoring tweaks to keep the broader-term searches stable.

Added

  • build-mule-integration — scripts/_suggest_nearest.py, a fuzzy nearest-match helper invoked by validate_before_build.sh to suggest the closest allowed error-type when the user's namespace:id miss has no exact match. Reduces time-to-fix on Cluster D validation failures.
  • build-mule-integration — scripts/.gitattributes to keep shell script line endings stable across contributor platforms.

[1.1.0] - 2026-05-18

Added

  • develop-pdk-policy — new skill that drives the full lifecycle of a custom Flex Gateway policy with the Policy Development Kit (PDK): prerequisite checks, anypoint-cli-v4 pdk policy-project create, make setup / build-asset-files / build, local execution via the scaffolded playground/ (make run against a Dockerized Flex Gateway in local-disconnected mode), then make publish and make release to Anypoint Exchange. Includes an upgrade-PDK runbook and troubleshooting for the most common toolchain failure modes. Lets agents take a developer from "I want a custom policy" through to a released Exchange asset without leaving the IDE.
  • pdk-templates — companion prose-only reference skill bundling 30 vetted, compilable PDK feature templates locally under templates/. Pulled from the upstream mulesoft-mcp-server mule-flex-pdk-service snapshots so the skill works offline with no MCP dependency. Covers JWT (validate + generate), OAuth2 introspection, header/body manipulation, body streaming, rate limiting, spike control, caching, distributed locks, worker variables, control flow, contracts, CORS, IP filtering, JSON/XML validators, outbound HTTP calls, gRPC, DataWeave evaluation, data storage, timers, logging, metadata, policy violations, stop_iteration, outbound-policy marker, and PDK unit testing setup. Multi-file bundles (grpc/, dataweave/, http_call/, stop_iteration/) ship as subdirectories with explicit destination guidance for each companion file (Cargo.toml.snippet, gcl.yaml, build.rs, proto/). Pairs with develop-pdk-policy, which owns scaffold/build/publish lifecycle.
  • pdk-unit — new skill that drives the unit-testing workflow for custom Flex Gateway PDK policies: deciding unit vs integration coverage, wiring src/tests/ (the scaffold ships tests/ for integration tests but not src/tests/ for unit tests), writing a first UnitTestBuilder test against crate::configure, factoring reusable TestConfig helpers, mocking HTTP upstreams via closures or TraceBackend capture, asserting on status / headers / PolicyViolation, and running make test / cargo test. Bundles six drop-in templates (hello test, config helper, upstream mock, trace-backend capture, violation assertion, src/tests/ module wiring) under templates/. Cross-links to pdk-templates/templates/unit_testing.md for the full pdk-unit API reference (no duplication) and to develop-pdk-policy for scaffold / build / publish lifecycle. Closes the testing gap left by those two skills.
  • pdk-test — new skill that drives the integration-testing workflow for custom Flex Gateway PDK policies: scaffolding tests/ with common/ helpers, writing RequestBuilder + assert_response! tests against a real Flex Gateway instance via make run, handling multi-request flows, testing configuration variants, and debugging test failures with RUST_LOG and Docker log inspection. Bundles templates for test structure and common patterns.

Changed

  • package.json files array now includes */templates/** (added alongside */references/**) so the bundled PDK templates ship in the published tarball.

[1.0.4] - 2026-05-18

Removed

  • build-mule-integration — dropped the mule-http-connector:1.11.2 → 1.11.1 pin in scripts/get_latest_connector.sh. The 1.11.2 POM has been republished on Exchange with the correct <parent> and <dependencies>, so the workaround is no longer needed and get_latest_connector.sh now passes through whatever Exchange returns.

[1.0.3] - 2026-05-14

Changed

  • build-mule-integration — synced with the upstream agent-evaluation lab (v12 of the skill).
    • Surfaces private (UUID-groupId) connectors published to a customer's Exchange tenant as first-class candidates alongside public connectors. The get_latest_connector.sh ranking already returned these rows; the prose now tells the agent to treat them as real options instead of noise, and to escalate via AskUserQuestion when both a public and a private connector match the same system family.
    • Step 3 "Common search terms" table rewritten with broader system names (salesforce, database, http, netsuite, servicenow, jms, slack) instead of narrow mule-<name>-connector strings, so private assets whose assetId does not share tokens with the public connector still surface.
    • New "Term breadth" guidance under the mandatory-search rule, plus updates to "No HTTP fallback without evidence" explaining UUID-format groupIds.
    • Step 16 gains a pre-mvn static validator (scripts/validate_before_build.sh) that checks the connector error-type whitelist (Cluster D), namespace ↔ pom.xml dependency parity (Cluster A2-A5), and canonical XSD URL shape — fast line-numbered diagnostics instead of a 30 s+ Maven failure.
    • scripts/describe_connector.sh now caches per-connector and per-operation errorTypes to tmp/connector-errors/, which the new validator reads.
    • scripts/get_latest_connector.sh ranking/scoring tweaks to keep the broader-term searches stable.

Fixed

  • package-lock.json was pinned to 1.0.1 while package.json had moved to 1.0.2; the lock file is now regenerated and consistent with the current package version.

[1.0.2] - 2026-05-14

Fixed

  • Corrected the spelling of @salesforce/mulesoft-vibes-skills in package metadata.

[1.0.1] - 2026-05-12

Added

  • repository field added to package.json so the published npm package links back to this repo.

Fixed

  • release-skills workflow and an earlier package-name typo.

[1.0.0] - 2026-05-08

Added

  • Initial public release of @salesforce/mulesoft-vibes-skills with the following skills:
    • build-mule-integration
    • create-project-template
    • create-mule-run-config / update-mule-run-config / delete-mule-run-config / execute-mule-run-config
    • generate-doc-description
    • run-system-diagnostics
    • secure-mule-app
  • npm publish workflow under .github/workflows/release-skills.yml.