feat(agentconfig-mcp): WeveNova planner MCP for planner-skill integration - #253
Open
Harsheet jain (jainharsheet77) wants to merge 4 commits into
Conversation
Harsheet jain (jainharsheet77)
force-pushed
the
harsheetjain/planner-skill-wevenova-mcp
branch
from
August 26, 2026 13:49
5b57e51 to
5821392
Compare
| Flow-2 "what am I assigned?" file. | ||
| the phase files (research, interview, model, assign, evaluate, capture), the sync | ||
| file (pull/push the plan with the shared planner), and the Flow-2 "what am I | ||
| assigned?" file. |
There was a problem hiding this comment.
For queries like what am I assigned reach out to mcp and syc as that returns tasks only for the roles and there is no api taht exposes what roels do I have so filtering is intrinsic to call for tasks with $filter
| interview or re-asking the objective. Start over only on explicit confirmation. | ||
| **Before interviewing, check for an existing plan — and pull first.** The planner | ||
| invisibly pulls the shared plan on entry; if a plan then exists (freshly pulled or | ||
| already at `workspace/plan/plan.json`), resume it — show its latest state and the |
There was a problem hiding this comment.
Now there can be many plans as we can have multiple plans in wevenova, get all the plans the person has access to and list and ask which pan would they like to resume, IT is fine to not choose any and create a new one altogether.
…egration Copy the self-contained WeveNova agent-configuration MCP package onto the planner-skill branch so the planner skill can drive WeveNova projects, plans, tasks, and role attestations via the ess-planner MCP server during integration testing. Ported verbatim from harsheetjain/wevenova-mcp-for-planner (PR microsoft#251). Tests: tests/mcp/agentconfig planner suites, 40 tests all passing. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Make the shared planner the source of truth for the /planner experience: the local plan.json becomes a cache that is pulled on entry and pushed as one object on create. The CLI stays network-free; the skill orchestrates the tool calls and shuttles JSON through the new sync seam. - plan_model: add the sync-seam mirror fields (configuringAgentName + etag/ syncedAt on the plan, remoteId/etag on each task), a CONFIGURING_AGENT_NAMES vocab, an acceptanceCriteria context group, and set_configuring_agent_name / set_remote_identity; back-fill + validate the new fields. - sync: new pure mapping module (no network, no IO) — to_remote_plan_body / to_remote_task_body export (only allowed fields; acceptance promotion; assignee flattening) and hydrate_from_remote import (Blocked<->Cancelled and acceptance bridges, Principal reconstruction, case-insensitive reads, outputs list->dict), plus stamp_remote_ids. - cli: add set-agent-name, export-remote-plan, import-remote-plan, stamp-remote. - skill: rewire SKILL.md to pull-first on entry and publish-as-one-object + activate on create; add sync.md; answer "what am I assigned?" from the service's caller-filtered tasks. Never name the backend in skill prose. - tests: tests/planner/test_sync.py covers export shape, the vocab bridges, the hydrate round-trip, and the four new CLI commands. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
…ndency-ordered tasks Adds the /roles skill: attest a person into an ESS rollout role so the shared plan's role-pooled tasks become visible to them. Backed by a directory name->object-id resolver (scripts/roles/cli.py resolve-person) that reuses the kit's Graph integration; role reads/writes stay planner-tool calls. Also renders plan task lists in execution order (a task producing an artifact another consumes is listed first) via a stable, cycle-tolerant topological sort in plan_model.ordered_tasks(), used by summary and the 'what am I assigned?' view; and nudges attestable-role assignment at plan creation. Skill prose stays backend-agnostic; routing, menu, and the planner assign nudge hook are wired. Tests: tests/roles resolver, tests/flightcheck search_users, and plan_model ordering. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Harsheet jain (jainharsheet77)
force-pushed
the
harsheetjain/planner-skill-wevenova-mcp
branch
from
August 26, 2026 14:54
d7ab659 to
372d022
Compare
…block Archived-via-update, normalize task keys
- _mutate_with_etag_recovery: never blind-replay after a 412. An advanced ETag
is indistinguishable from a concurrent edit, so raise an actionable "entity
changed, re-read and reapply" 412 instead of clobbering another writer.
Verified live: a stale If-Match PATCH returns 412 and the prior value is
preserved (no replay).
- client._request: surface (don't replay) an ambiguous 502/503/504 or network
error on a non-idempotent request (an unkeyed create) so a
committed-but-unacknowledged POST can't be duplicated; reads, keyed
mutations, and PATCH/DELETE keep bounded retry, and 429 always retries.
- update_project_plan: reject {"status":"Archived"} and route archiving (which
cancels the plan's tasks) through the destructive archive_project_plan tool.
Verified live: archiving cancels the plan's tasks.
- update_project_plan_task: normalize patch keys to canonical camelCase and
reject a field named twice under different casing. Verified live: the surface
deserializes case-insensitively, so this is deterministic-shape hardening,
not a required fix -- the "wrong casing is rejected" premise does not hold.
- server_planner: reconcile tool docstrings/instructions with the no-replay 412.
- tests: pin no-replay 412 (task/plan/delete), Archived rejection, key
normalization + duplicate rejection, and unkeyed-create retry suppression.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Brings the self-contained WeveNova AgentConfiguration MCP (the
ess-plannerserver) onto
feature/planner-skillso the planner skill can drive WeveNovaprojects, plans, tasks, and plan role attestation end to end during
integration testing.
Single commit; the change set is MCP-only (9-file package under
solutions/ess-maker-skills/src/mcp/agentconfig/+ 3 planner test suites undertests/mcp/agentconfig/).Tools
server_planner.pyexposes 23 MCP tools over the beta surface — projects(list/get/create/archive), plans (list/get/create/update/activate/archive),
tasks (list/get/create/role-assign/update/set-state/complete/delete), and role
attestation (list/attest/list-assignments/get/revoke). Identity and tenant are
always taken from the access token, never tool arguments.
Self-healing conflict recovery
Every
If-Matchmutation now routes through_mutate_with_etag_recovery, so aplanner agent does not have to hand-roll optimistic-concurrency bookkeeping:
mutation once with the fresh ETag — but only when the version actually
moved, so a genuine lost update is never silently clobbered. WeveNova bumps a
task's version as a side effect of ledger reconciliation (e.g. completing a
producer task reconciles an artifact a consumer task references), so an ETag a
caller just read can go stale through no edit of its own.
and, when it is not Active, replace the backend's generic conflict text with an
actionable "activate the plan first" message.
Validation
(projects/plans/tasks/roles, full task lifecycle with ETag threading, multi-active
plans). All tools pass; the only non-pass observed was backend test-data
saturation (
MaxPlansPerProject=10), not an MCP defect.tests/mcp/agentconfig— 46 passing (request-construction contract viahttpx.MockTransport, static + protocol tool-surface contract, and six newrecovery tests covering the 412 retry, the "version unchanged ⇒ re-raise"
guard, and the 409 activate-first clarifier).
Notes
ess-plannerMCP is environment-configured (AGENTCONFIG_PROJECTS_BASE_URL,AGENTCONFIG_ACCESS_TOKEN_FILE); it does not depend on.vscode/mcp.defaults.json.test_client,test_server_contract,test_widget_protocol) are intentionally omitted — they assert onlanding-page-config artifacts not present on
feature/planner-skill.