AgentXchain v2.113.0

Released: 2026-04-17

Highlights

Mission decomposition — agentxchain mission plan generates dependency-ordered workstreams from a mission goal via LLM-assisted planning, with strict schema validation and durable plan artifacts
Plan approval gate — mission plan approve enforces latest-only governance with deterministic supersession lineage
One-command workstream launch — mission plan launch executes immediately through the chain runner with preallocated chain IDs, not deferred bookkeeping
Dashboard plan visibility — Mission view surfaces latest plan detail, workstream launch status, launch records with chain linkage, and prior plan revisions via dedicated GET /api/plans endpoint
Missions docs update — /docs/missions now covers the full decomposition flow end-to-end

New agentxchain mission plan command generates a structured decomposition of a mission goal into dependency-ordered workstreams:

LLM-assisted one-shot generation with operator-supplied --constraint and --role-hint options
Goal derived from the mission artifact — no split-brain goal drift
Strict schema validation: required fields, type checking, duplicate ID detection, dependency reference validation, chain_id pre-allocation rejection
Durable plan artifacts under .agentxchain/missions/plans/<mission_id>/<plan_id>.json
launch_status derived from dependency edges at creation time (ready for no dependencies, blocked for non-empty depends_on)
Revision by supersession — new plans auto-set supersedes_plan_id to the prior plan's ID
mission plan show and mission plan list for plan inspection

Latest-only governance: only the newest proposed plan may transition to approved
Approving a newer plan supersedes any older active proposed or approved plans
Fails closed on: stale targets, already-approved plans, non-proposed plans
Deterministic recency — plan lineage and latest-plan resolution are monotonic, not random-suffix-dependent

One-command execution surface, not deferred bookkeeping
Requires --workstream <id> and an approved plan
Preallocates chain_id at launch, executes immediately via executeChainedRun
Dependency satisfaction based on latest run status in bound chain report — single-run chains with terminal_reason: "chain_limit_reached" correctly unblock dependents
Mission attachment happens after chain report exists, not before
Five workstream launch states: ready, blocked, launched, completed, needs_attention
On failure: workstream sets needs_attention, plan sets needs_attention, dependents remain blocked

Dedicated GET /api/plans endpoint with optional ?mission=<id> filter and ?limit=N pagination
Integrated into Mission dashboard view: latest plan with workstream table, launch records with chain linkage, status breakdown, previous plans
Recursive file watching for missions/plans/**/*.json — plan changes invalidate both /api/plans and /api/missions
Empty-state guidance: "Run agentxchain mission plan <mission_id> to generate one."

/docs/missions now documents both operator paths: direct chained execution and decomposed planning
Covers plan artifacts, GET /api/plans, launch-state visibility, and immediate-execution boundary
Cross-linked from CLI docs and run-chaining guide