@chris1807/claude-kit 2.1.22 → 2.1.23

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@chris1807/claude-kit",
-  "version": "2.1.22",
+  "version": "2.1.23",
   "description": "Claude Code starter kit for Azure DevOps teams — agents, hooks, MCP servers, slash commands, and end-to-end work item → PR → release → deploy workflow automation",
   "type": "module",
   "bin": {

package/templates/agents/global/manager.md

@@ -67,6 +67,12 @@ When tasks are independent, delegate in parallel:
 - Build validation and lint checking can run simultaneously
 - Frontend work depends on backend APIs being defined (but not necessarily implemented)
 
+Issue independent delegations as multiple `Task` calls in a single message so they run concurrently.
+
+### Ultracode is not yours to run
+
+**Ultracode** (the `Workflow` tool — deterministic fan-out across many agents) is a *main-loop* capability. You run as a subagent under `Task` and do **not** have the `Workflow` tool, so never assume you can author a workflow. Your parallelism is limited to issuing several `Task` delegations at once (above). If a job genuinely calls for ultracode-scale fan-out — e.g. one agent per work item across a whole backlog — that belongs to the main loop or a slash command; surface the suggestion to the user rather than attempting it yourself.
+
 ## How to Start a Feature
 
 When the user says "implement Feature X" or "work on [feature]":
@@ -111,6 +117,8 @@ The user may invoke slash commands directly instead of asking you to orchestrate
 | `/status release 24` | Check release, pipeline, or work item status |
 | `/review 142` | Code review a PR |
 | `/cleanup-branches` | Delete merged branches |
+| `/close-orphan-tasks` | Close open Tasks whose parent has shipped (Ready to Deploy / Deployed / Closed) |
+| `/plan-backlog` | Sweep the backlog for Dev Ready stories and propose child tasks |
 
 If the user asks you to "deploy to staging" or "create a release", suggest the appropriate slash command.
 

package/templates/commands/plan-backlog.md

@@ -89,6 +89,15 @@ Continue? (yes / cancel)
 
 **Wait for the user.** If `cancel`, stop with no changes.
 
+### Ultracode mode (optional fan-out)
+
+Only when **ultracode is on** (a system-reminder confirms it, or the user typed `ultracode`): the analysis in 5a–5c is independent per story, so pre-compute all proposals in parallel with the `Workflow` tool instead of analyzing one story at a time.
+
+- Fan out **one agent per story** that does 5a–5c — re-read the story, map points → budget, tailor the task list — and returns a structured proposal (story id, points, budget, and the list of `{title, hours}` tasks). Use a `schema` so each agent returns validated JSON.
+- Then run Step 5's loop **using the pre-computed proposals** — but keep 5d (approval) and 5e (creation) exactly as written: present each proposal, wait for `yes / edit / skip / cancel-all`, and create tasks only after approval. **Never fan out the approval or the work-item creation** — those stay sequential and interactive.
+
+If ultracode is off, ignore this and run Step 5 the normal sequential way. The output is identical either way; ultracode only makes the analysis faster for large backlogs.
+
 ## Step 5: Per-Story Task Breakdown (loop)
 
 For each remaining story, in order:

package/templates/infrastructure/CLAUDE-WORKFLOW.md

@@ -29,6 +29,25 @@ You (give task)
   └── reviewer → reviews code quality
 ```
 
+### Ultracode (multi-agent workflows)
+
+The **agent pipeline** above (`manager` → specialists) is sequential delegation: one agent at a time, through the `Task` tool. **Ultracode** is a different, harness-level gear — it authorizes the `Workflow` tool to fan out many agents in parallel under a deterministic script. The two compose; ultracode does not replace the pipeline, it parallelizes the parts of it that are embarrassingly parallel.
+
+Ultracode is **opt-in**. It is on only when a system-reminder confirms it, when you type `ultracode` in a prompt, or when a slash command's instructions say to use `Workflow`. When it is off, everything below runs the normal sequential way — nothing changes.
+
+**Who can call `Workflow`:** the main Claude Code loop, and slash commands (they expand into the main conversation). **Subagents cannot** — so the `manager` agent, running under `Task`, never authors a workflow. Ultracode-scale fan-out is a main-loop / slash-command concern.
+
+**Kit operations that benefit from ultracode (when it's on):**
+
+| Operation | Fan-out unit |
+|-----------|--------------|
+| `/plan-backlog` | one agent per Dev Ready story — analyze, point, and propose tasks in parallel |
+| Backlog / board audits | one agent per work item — find stale, mislabeled, orphaned, or unestimated items |
+| Multi-file or cross-layer review | one agent per file/dimension, then adversarial verify before reporting |
+| Repo-wide sweeps (rename, dependency bump, pattern migration) | one agent per site, worktree-isolated |
+
+Short, single-query operations (`/status`, `/explain`, `/close-orphan-tasks`) do **not** need ultracode — they are already one pass and gain nothing from fan-out. Reach for `Workflow` when the work-list is large and the per-item work is independent.
+
 ### Sensitive Data Policy
 
 **NEVER query, display, or expose sensitive PII fields from the database — even if the values are encrypted.** This includes TIN, SSN, EIN, TaxId, BankAccountNumber, RoutingNumber, and any `Encrypted*` variants. Even encrypted/hashed values must not appear in output, logs, or summaries. When querying collections that may contain sensitive fields, always use explicit inclusion projections listing only the non-sensitive fields needed. If a user requests access to sensitive data, direct them to use the application UI.