@uipath/skills 1.195.0 → 1.196.0

package/commands/install-permissions.md

@@ -60,14 +60,54 @@ Split by risk. `allow` = read-only or local-only commands; `ask` = commands with
       "Bash(uip flow edge *)",
       "Bash(uip solution new *)",
       "Bash(uip solution project *)",
-      "Bash(uip solution resource *)",
+      "Bash(uip solution resources *)",
       "Bash(uip agent init)",
       "Bash(uip agent init *)",
       "Bash(uip agent validate *)",
 
       // RPA skill — build, get-errors, find-activities, create-project, inspect-package, etc.
       // All local; no cloud side effects (publish goes through `uip solution upload/publish`).
-      "Bash(uip rpa *)"
+      "Bash(uip rpa *)",
+
+      // uipath-troubleshoot — read-only diagnostics. Scoped PER VERB on purpose:
+      // a bare `uip or jobs *` would also allow stop/start/restart/resume;
+      // `uip resource *` would allow delete/upload/add; `uip maestro bpmn instance *`
+      // would allow cancel/retry/goto/variables-set. Mutating verbs are left out so
+      // they keep prompting (fails closed — a new CLI verb is never auto-allowed).
+      "Bash(uip or jobs get *)",
+      "Bash(uip or jobs list *)",
+      "Bash(uip or jobs logs *)",
+      "Bash(uip or jobs history *)",
+      "Bash(uip or jobs traces *)",
+      "Bash(uip or jobs healing-data *)",
+      "Bash(uip or assets get *)",
+      "Bash(uip or assets list *)",
+      "Bash(uip or licenses info *)",
+      "Bash(uip or licenses list *)",
+      "Bash(uip or users get *)",
+      "Bash(uip or users list *)",
+      "Bash(uip resource assets get *)",
+      "Bash(uip resource assets list *)",
+      "Bash(uip resource queue-items get *)",
+      "Bash(uip resource queue-items get-history *)",
+      "Bash(uip resource queue-items get-last-retry *)",
+      "Bash(uip resource queue-items list *)",
+      "Bash(uip resource queues get *)",
+      "Bash(uip resource queues list *)",
+      "Bash(uip docsai ask *)",
+      "Bash(uip traces spans get *)",
+      "Bash(uip maestro bpmn instance get *)",
+      "Bash(uip maestro bpmn instance list *)",
+      "Bash(uip maestro bpmn instance incidents *)",
+      "Bash(uip maestro bpmn instance element-executions *)",
+
+      // uipath-troubleshoot — investigation workspace (local, throwaway) + bundled playbooks.
+      // Write/Edit prompt by default; the Read rules are insurance (file reads are not gated
+      // by Claude Code, but they document intent and survive stricter permission modes).
+      "Read(.local/investigations/**)",
+      "Write(.local/investigations/**)",
+      "Edit(.local/investigations/**)",
+      "Read(//**/uipath-troubleshoot/references/**)"
     ],
     "ask": [
       "Bash(uip login)",
@@ -95,3 +135,7 @@ Split by risk. `allow` = read-only or local-only commands; `ask` = commands with
 - The full variant's `ask` block intentionally guards `uip solution upload`, `uip flow debug`, `uip flow pack`, `uip solution publish`, and `uip login` — these produce real cloud side effects (publishing, executing flows, tenant auth). The `uipath-maestro-flow` skill's rules explicitly require user consent before cloud-executing a flow.
 - `uip agent init` is in `allow` (scaffolding — same as `uip flow init`). It does not publish to the cloud; it creates local agent project files.
 - `uip rpa *` covers the full RPA subcommand tree (`build`, `get-errors`, `create-project`, `find-activities`, `inspect-package`, `focus-activity`, etc.). None of these publish to the cloud — RPA deployment flows through `uip solution upload` / `publish` (already guarded in `ask`).
+- The `uipath-troubleshoot` skill is read-only — it inspects jobs, assets, queue-items, traces, and Maestro instances but never mutates them. Its verbs are allowlisted **per read verb** (e.g. `uip or jobs get *`, not `uip or jobs *`) so mutating siblings keep prompting: `or jobs stop/start/restart/resume`, `or assets create/delete/update/share`, `or licenses toggle`, `resource … delete/upload/add/update`, and `maestro bpmn instance cancel/retry/goto/pause/resume/migrate/message send/variables-set`. This fails closed — a newly shipped CLI verb is never auto-allowed.
+- The troubleshoot rules cover the canonical short form `uip or …`. If the agent uses the long-form alias `uip orchestrator …`, it will still prompt; the fix is to keep the skill on the short form, not to widen the allowlist.
+- `Read/Write/Edit(.local/investigations/**)` cover the throwaway investigation workspace the troubleshoot sub-agents write to (`state.json`, `hypotheses.json`, `evidence/`, `raw/`, `depth-check.json`, etc.). Without the `Write`/`Edit` rules every investigation step prompts.
+- `Read(//**/uipath-troubleshoot/references/**)` matches the skill's bundled playbooks wherever the plugin is installed (`//` anchors to the filesystem root; `**/` matches arbitrary intermediate dirs). Per the Claude Code docs, file reads are never gated, so this rule is insurance rather than load-bearing.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@uipath/skills",
-  "version": "1.195.0",
+  "version": "1.196.0",
   "description": "UiPath agent skills for Claude Code, Codex, Cursor, Copilot, Gemini and OpenCode — RPA, UI automation, UI testing, coded agents/apps/workflows, and troubleshooting. Distributed as the UiPath Claude Code plugin.",
   "author": {
     "name": "UiPath"

package/skills/uipath-admin/SKILL.md

@@ -1,13 +1,11 @@
 ---
 name: uipath-admin
-description: "UiPath Admin via `uip admin` — Identity Server (users, groups, robot accounts, external OAuth2 apps, secrets), Authorization (custom roles, role assignments, permission catalog, effective-access via check-access PDP), OMS (org read/update, tenant lifecycle, service provisioning, regions, async operation polling), IP Restriction (allowlist, enforcement switch, bypass rules, lockout safety), Audit (event sources, paginated queries, ZIP exports — login history, compliance dumps, who-did-what-when-where on a resource). For Orchestrator-specific roles/permissions/folders/jobs→uipath-platform. For RPA workflows→uipath-rpa."
+description: "UiPath Admin via `uip admin` — Identity Server (users, groups, robot accounts, external OAuth2 apps, secrets), Authorization (custom roles, role assignments, permission catalog, effective-access via check-access PDP), OMS (org read/update, tenant lifecycle, service provisioning, regions, async operation polling), IP Restriction (allowlist, enforcement switch, bypass rules, lockout safety), Audit (event sources, paginated queries, ZIP or single-CSV exports — login history, compliance dumps, who-did-what-when-where on a resource). For Orchestrator-specific roles/permissions/folders/jobs→uipath-platform. For RPA workflows→uipath-rpa."
 allowed-tools: Bash, Read, Write, Edit, Glob, Grep, AskUserQuestion
 ---
 
 # UiPath Admin
 
-> **Preview** — Under active development. Command coverage will expand.
-
 Administrative operations on UiPath via `uip admin` — Identity Server, Authorization, OMS, IP Restriction, Audit. Per-area workflows, command references, and procedures are in the linked files below — this file is the entry contract.
 
 ## When to Use This Skill
@@ -45,7 +43,7 @@ Administrative operations on UiPath via `uip admin` — Identity Server, Authori
 - **Manage IP allowlisting** — add / update / delete CIDR entries that gate inbound access
 - **Toggle IP-restriction enforcement** — turn the org-wide allowlist switch on or off (with lockout safety)
 - **Manage bypass rules** — URL-pattern exceptions to IP allowlisting
-- **Look up the caller's public IP** — sanity check before enabling enforcement
+- **Look up the caller's public IP** — answer "what's my IP?" / "what public IP does the platform see for me?" directly with `ip-restriction my-ip`; also the safety pre-flight before enabling enforcement
 
 ### Audit
 
@@ -53,7 +51,7 @@ Activate on both **explicit audit requests** and **natural-language investigatio
 
 - **Explicit** — `uip admin audit` commands; list sources / targets / types; query, filter, paginate, or export events; CSV/ZIP dump of audit history for a window.
 - **Query audit events** — list event sources, filter events by source / target / type / user / status / time window at org or tenant scope
-- **Export audit events** — chunked ZIP download from the long-term store, per UTC day, with atomic abort on any chunk failure
+- **Export audit events** — chunked download from the long-term store (one call per UTC day, atomic abort on any chunk failure) as a ZIP of day-wise JSON files (default) or a single merged CSV via `--file-format csv`
 - **Membership / license phrasings** — "who joined / left the organization", "who was made an admin", "license changes", "cross-tenant audit"
 - **Sign-in / authentication phrasings** — "failed/successful logins", "login history for user X", "who's been signing in"
 - **Tenant-activity phrasings** — "what happened on tenant X", "asset/queue/folder edits", "queue items processed", "job failures", "Action Center task changes", "Apps / AgentHub / Document Understanding / Integration Service / Test Manager activity"
@@ -102,14 +100,14 @@ Each rule is the agent contract. Per-area detail is in the linked reference file
 
 ### Audit
 
-23. **Disambiguate `org` vs `tenant` scope before querying.** If the prompt is vague AND no prior turn fixed the scope, **stop and ask once** — never silently default to `tenant`. Routing table (user-phrasing → scope + why it lives where) and investigation playbooks: [audit-workflow-guide.md → Audit scope disambiguation](references/audit-workflow-guide.md#audit-scope-disambiguation--route-by-user-phrasing).
+23. **Disambiguate `org` vs `tenant` scope before querying.** If the prompt is vague AND no prior turn fixed the scope, **stop and ask once** — never silently default to `tenant`. **Scope is a positional subgroup, NOT a flag:** write `uip admin audit org sources` / `uip admin audit tenant events` — there is no `--scope` option (`audit sources --scope organization` is invalid). Routing table (user-phrasing → scope + why it lives where) and investigation playbooks: [audit-workflow-guide.md → Audit scope disambiguation](references/audit-workflow-guide.md#audit-scope-disambiguation--route-by-user-phrasing).
 24. **`audit <scope> events` returns an object, not a bare array.** Shape is `{auditEvents, next, previous}`. Do not index `Data[0]`; read `Data.auditEvents[]`. **Cursor semantics are chronological**: `next` = newer events, `previous` = older events. The default newest-backward walk follows `previous`.
 25. **`--limit` paginates internally — never loop on `--from-date` / `--to-date` to "paginate".** The server clamps `maxCount` to `[10, 200]` per request; when the user wants more than 200, the CLI fetches `ceil(N/200)` pages under the hood. Pass `--limit 500` (or larger) — do NOT re-implement pagination in the agent.
 26. **Discover via `audit <scope> sources` first — never invent source / target / type GUIDs.** The catalog response gives the GUIDs you pass to `events --source / --target / --type`.
 27. **Bound the time window, ISO 8601 in UTC.** Don't call `audit <scope> events` without `--from-date` and `--to-date` on a noisy tenant. Accepted formats: date-only (`2026-04-01`) or with time (`2026-04-01T14:30:00Z`). **`--to-date` is inclusive of the exact instant** — to capture a full final day, pass the start of the next day or `T23:59:59.999Z`.
 28. **`--tenant-id` is silently ignored on `org`-scoped audit commands.** If you find yourself reaching for it on `audit org events`, switch to `audit tenant` instead.
 29. **On 401 from audit, do NOT retry.** The token is missing the `Audit.Read` scope; tell the user to `uip logout && uip login`.
-30. **`audit <scope> export` writes a ZIP from the long-term store.** `--from-date`, `--to-date`, and `--output-file` are all required; dates per Rule 27. **Never overwrite a path the user did not explicitly approve** — surface the resolved `--output-file` and confirm before running.
+30. **`audit <scope> export` writes a ZIP (default) or a single merged CSV from the long-term store.** `--from-date`, `--to-date`, and `--output-file` are all required; dates per Rule 27. **`--file-format <zip|csv>`** selects the shape: `zip` (default) is one JSON file per UTC day; `csv` merges every event into one CSV — pick `csv` when the user wants a flat spreadsheet/Excel-friendly dump and `zip` for archival or per-day JSON. Match the `--output-file` extension to the format (`.zip` / `.csv`). **Never overwrite a path the user did not explicitly approve** — surface the resolved `--output-file` and confirm before running.
 
 ### IP Restriction
 
@@ -146,6 +144,7 @@ One row per common goal. Per-area workflows are in the reference files.
 | **See what a principal can do** | `uip admin authorization check-access <USER_GUID_OR_EMAIL> --scope <Tenant\|Folder> --output json` (Rule 15) |
 | **Create a tenant** | [tenant-management.md](references/tenant-management.md) — region + default-services resolution, file-body shape, operation polling (Rule 18) |
 | **Add a tenant service** | `tenants services list-available --region <R>` → `tenants services add --tenant-id <TID> --service <SVC>` (verify post-state per Rule 22) |
+| **Look up my public IP** ("what's my IP?", "what public IP does the platform see for me?") | `ip-restriction my-ip --output json` → return `Data.ipAddress`. Standalone — no enforcement context required |
 | **Enable IP allowlist enforcement** | `ip-restriction my-ip` → verify covered by `ip-ranges list` → `ip-restriction enforcement enable --confirm` (Rule 31) |
 | **Query audit events / export** | [audit-workflow-guide.md](references/audit-workflow-guide.md) — scope disambiguation + 4 investigation playbooks (who-did-X, login history, date-range dump, overview) |
 

package/skills/uipath-admin/references/audit-commands.md

@@ -18,7 +18,7 @@ uip admin audit
     └── export
 ```
 
-`org` and `tenant` are **subject subgroups**. Same three verbs under each. The two trees are 100% verb-symmetric — any flag valid on `tenant events` is also valid on `org events` (except `--tenant-id`, which is tenant-only).
+`org` and `tenant` are **subject subgroups** — pass scope as this **positional segment** (`uip admin audit org sources`, `uip admin audit tenant export`), **never as a `--scope` flag** (there is no `--scope` option; `audit sources --scope organization` is invalid). Same three verbs under each. The two trees are 100% verb-symmetric — any flag valid on `tenant events` is also valid on `org events` (except `--tenant-id`, which is tenant-only).
 
 ## Output `Data` shape varies by verb — quick reference
 
@@ -26,7 +26,7 @@ uip admin audit
 |---|---|
 | `audit <scope> sources` | array of `AuditEventSourceDto` |
 | `audit <scope> events` | object `{auditEvents, next, previous}` |
-| `audit <scope> export` | object `{Path, Bytes, Format, Days, NonEmptyDays}` |
+| `audit <scope> export` | object `{Path, Bytes, Format, Days, NonEmptyDays}` (+ `Events` when `--file-format csv`) |
 
 `events` is the one verb that legitimately returns an object — pagination cursors live alongside the rows. Full shape detail per verb in the sections below.
 
@@ -138,31 +138,42 @@ uip admin audit tenant events --from-date 2026-04-22T00:00:00Z --to-date 2026-04
 
 ## uip admin audit `<scope>` export
 
-Stream a ZIP from the long-term audit store covering `[--from-date, --to-date]`.
+Export the long-term audit store covering `[--from-date, --to-date]` — as a ZIP of day-wise JSON files (default) or a single merged CSV.
 
 ```bash
+# Default — ZIP of one JSON file per UTC day
 uip admin audit tenant export \
   --from-date 2026-01-01 \
   --to-date 2026-02-01 \
   --output-file ./audit-jan.zip \
   --output json
+
+# Single merged CSV of every event
+uip admin audit tenant export \
+  --from-date 2026-01-01 \
+  --to-date 2026-02-01 \
+  --file-format csv \
+  --output-file ./audit-jan.csv \
+  --output json
 ```
 
 **Flags:**
 
 | Flag | Required | Description |
 |---|---|---|
-| `--output-file <path>` | **yes** | Where to write the ZIP. Parent dir is created if missing; existing file is overwritten. Resolved to absolute internally. |
+| `--output-file <path>` | **yes** | Where to write the export. Parent dir is created if missing; existing file is overwritten. Resolved to absolute internally. Match the extension to `--file-format` (`.zip` / `.csv`). |
 | `--from-date <iso>` | **yes** | Start of time interval. Both bounds are required by Commander before any HTTP call. |
 | `--to-date <iso>` | **yes** | End of time interval. |
+| `--file-format <zip\|csv>` | no | Output shape. `zip` (default) = one JSON file per UTC day inside a ZIP. `csv` = every event merged into a single RFC 4180 CSV under a shared header. Invalid values fail before any HTTP call with `Invalid --file-format '<v>'. Use 'zip' or 'csv'.` |
 | `--login-validity <minutes>` | no | Token-refresh hint. |
 | `--tenant-id <guid>` | no | **Tenant scope only.** Override the active tenant. |
 
 **Output `Code`:** `AuditOrgExport` / `AuditTenantExport`.
 
-**Output `Data`:**
+**Output `Data`:** `Format` echoes the chosen `--file-format`. The CSV path adds `Events` (total rows written, excluding the header); the ZIP path omits it.
 
 ```json
+// --file-format zip (default)
 {
   "Path": "C:\\absolute\\path\\to\\audit-jan.zip",
   "Bytes": 1841,
@@ -170,14 +181,25 @@ uip admin audit tenant export \
   "Days": 31,
   "NonEmptyDays": 27
 }
+
+// --file-format csv
+{
+  "Path": "C:\\absolute\\path\\to\\audit-jan.csv",
+  "Bytes": 98765,
+  "Format": "csv",
+  "Days": 31,
+  "NonEmptyDays": 27,
+  "Events": 1234
+}
 ```
 
 **Implementation notes (worth knowing for diagnostic conversations):**
 
-- The CLI issues **one HTTP call per UTC day** inside `[from, to]` and aggregates the per-day responses into a single output ZIP. Mirrors the `audit-dowload-from-longterm-store.sh` pattern in the AuditService repo.
-- Per-day entries land at the **root of the output ZIP** named `<YYYY-MM-DD>.txt` (no folder prefix). Each `.txt` is a JSON array of events with PascalCase keys (`Id`, `CreatedOn`, `EventType`, …).
-- On any single-day HTTP failure, **no file is written** and the error message identifies which day failed. Earlier successful chunks are not preserved.
-- `Days` reports the total number of UTC days requested; `NonEmptyDays` reports how many actually had events. A long export with `NonEmptyDays: 0` means the window was entirely idle, not that the export failed.
+- Both formats share the same fetch: the CLI issues **one HTTP call per UTC day** inside `[from, to]` and aggregates the per-day responses. Mirrors the `audit-dowload-from-longterm-store.sh` pattern in the AuditService repo.
+- **ZIP (default):** per-day entries land at the **root of the output ZIP** named `<YYYY-MM-DD>.txt` (no folder prefix). Each `.txt` is a JSON array of events with PascalCase keys (`Id`, `CreatedOn`, `EventType`, …).
+- **CSV:** the same per-day JSON arrays are parsed and merged into **one** RFC 4180 CSV (CRLF line endings, header row first). Columns follow the long-term-store field order — `Id, CreatedOn, OrganizationId, TenantId, ActorId, ActorName, ActorEmail, ActorDetails, EventType, EventSource, EventTarget, EventDetails, Status, ClientInfo` — with any extra server fields appended (union across events) so no data is dropped. `Status` is the numeric enum (`0`=Success, `1`=Failure); nested objects (e.g. `ClientInfo`) are JSON-stringified into the cell. String cells beginning with `= + - @` (or TAB/CR) are prefixed with a single quote to neutralize spreadsheet formula injection.
+- On any single-day HTTP failure (or, for CSV, a day whose payload is not valid JSON), **no file is written** and the error message identifies which day failed. Earlier successful chunks are not preserved (atomic export).
+- `Days` reports the total number of UTC days requested; `NonEmptyDays` reports how many actually had data. A long export with `NonEmptyDays: 0` means the window was entirely idle, not that the export failed. For CSV, `Events: 0` yields a header-only file.
 
 ---
 
@@ -187,7 +209,7 @@ These appear on every command (not just audit) — set at the program level by t
 
 | Flag | Description |
 |---|---|
-| `--output <table\|json\|yaml\|plain>` | Format of the success/failure envelope on stdout. Defaults to `json`. The ZIP body of `export` is unaffected — `--output` only controls the metadata envelope. |
+| `--output <table\|json\|yaml\|plain>` | Format of the success/failure envelope on stdout. Defaults to `json`. The export file body (ZIP or CSV) is unaffected — `--output` only controls the metadata envelope, NOT `--file-format`. |
 | `--output-filter <jmespath>` | JMESPath query applied to the envelope before printing. Useful for `events`/`sources`. Less useful for `export` (envelope is small). |
 | `--log-level <debug\|info\|warn\|error>` | Logger threshold. Logs go to stderr. |
 | `--log-file <path>` | Redirect logs from stderr to a file. |

package/skills/uipath-admin/references/audit-workflow-guide.md

@@ -174,7 +174,7 @@ For each event:
 
 **User asks:** "Export everything for compliance for Q4." / "I need the full audit log for January for the security review." / "Pull last month's events for tenant X as a ZIP."
 
-**Approach:** `export` straight to a ZIP. No need to query `events` first unless the user wants a preview.
+**Approach:** `export` straight to a file. Default is a ZIP of day-wise JSON files; add `--file-format csv` for a single merged CSV when the user wants a flat, spreadsheet/Excel-friendly dump. No need to query `events` first unless the user wants a preview. Match the `--output-file` extension to the format (`.zip` / `.csv`).
 
 ### Step 1 — Confirm scope and window
 
@@ -183,13 +183,21 @@ If the user is ambiguous about scope, ask once. For compliance reviews, **both**
 ### Step 2 — Export
 
 ```bash
-# Tenant scope — most events
+# Tenant scope — most events (default ZIP of day-wise JSON files)
 uip admin audit tenant export \
   --from-date 2026-01-01 \
   --to-date   2026-02-01 \
   --output-file ./audit-tenant-2026-01.zip \
   --output json
 
+# Tenant scope as a single merged CSV (flat, Excel-friendly)
+uip admin audit tenant export \
+  --from-date 2026-01-01 \
+  --to-date   2026-02-01 \
+  --file-format csv \
+  --output-file ./audit-tenant-2026-01.csv \
+  --output json
+
 # Org scope — admin events (memberships, license, tenant lifecycle)
 uip admin audit org export \
   --from-date 2026-01-01 \
@@ -198,9 +206,11 @@ uip admin audit org export \
   --output json
 ```
 
-The CLI issues one HTTP call per UTC day under the hood and aggregates daily responses into a flat ZIP. `Days` and `NonEmptyDays` in the result tell you how many calendar days had events.
+The CLI issues one HTTP call per UTC day under the hood. ZIP aggregates the daily responses into a flat ZIP; CSV parses the same daily JSON and merges every event into one CSV. `Days` and `NonEmptyDays` in the result tell you how many calendar days had data; for CSV, `Events` reports the total row count.
+
+### Step 3 — Verify the export
 
-### Step 3 — Verify the ZIP
+**ZIP** — list the per-day entries:
 
 ```bash
 unzip -l ./audit-tenant-2026-01.zip
@@ -218,6 +228,15 @@ audit-tenant-2026-01.zip
 
 Each `.txt` is a JSON array of audit events with **PascalCase** keys (`Id`, `CreatedOn`, `OrganizationId`, `ActorId`, `ActorName`, `EventType`, …) — different from the camelCase shape returned by the live `events` endpoint. Note this in the user's hand-off if they're going to feed the dump into other tooling.
 
+**CSV** — inspect the header and row count:
+
+```bash
+head -1 ./audit-tenant-2026-01.csv          # shared header (PascalCase columns)
+python3 -c "import csv; print(sum(1 for _ in csv.reader(open('./audit-tenant-2026-01.csv'))) - 1, 'rows')"
+```
+
+One header row, then every event across all days as a data row (same PascalCase column names as the ZIP's JSON keys). The row count should match `Events` in the result envelope.
+
 Edge cases the CLI handles automatically — surface in your hand-off only if they appear:
 
 - **Nested ZIPs**: when a single day's response is itself a ZIP-of-ZIPs, inner files are renamed `<inner>_<outer>.txt` rather than collapsed.
@@ -282,14 +301,14 @@ Two or more signals? Run them in sequence and stitch the results in the final re
 - **`tenant` events without an active tenant fail loudly.** If `uip login` has no tenant selected, every tenant-scoped command throws. Either re-`uip login` and pick a tenant, or pass `--tenant-id <guid>` on every call.
 - **`events` cursor pagination is chronologically reversed from intuition.** `next` = newer (often null), `previous` = older (the typical "load more"). The CLI tool follows `previous` automatically when you bump `--limit > 200` — don't re-implement this in the agent.
 - **Date-only ISO strings are interpreted as UTC midnight.** `--from-date 2026-01-01` means `2026-01-01T00:00:00Z`. To capture the full final day in `--to-date`, use `2026-02-01` (exclusive next day) or `2026-01-31T23:59:59.999Z`.
-- **The export ZIP's per-day files are JSON, not CSV, and use PascalCase keys.** Different from the camelCase live `events` endpoint. Don't paste an export directly into a parser expecting the live shape.
+- **Export format depends on `--file-format`.** The default `zip` holds one **JSON** file per UTC day (not CSV) with **PascalCase** keys; `--file-format csv` produces a single merged **CSV** whose header uses those same PascalCase field names. Both differ from the camelCase live `events` endpoint — don't paste an export into a parser expecting the live shape. In the CSV, `Status` is numeric (`0`/`1`) and `ClientInfo` is a JSON-stringified cell.
 - **Org sources and tenant sources are different sets.** Don't reuse a GUID from `org sources` in a `tenant events` query — the filter will silently match nothing.
 
 ## Output Etiquette — after an audit query or export
 
 After every `events` or `export` call, surface the following before waiting for the user's next-step choice. Do not chain mutations.
 
-1. **Operation & result** — e.g. `Found 47 audit events on tenant T in the last 7 days` or `Wrote 123,456 bytes to /path/to/audit.zip (3 days, 2 non-empty)`.
+1. **Operation & result** — e.g. `Found 47 audit events on tenant T in the last 7 days`, `Wrote 123,456 bytes to /path/to/audit.zip (3 days, 2 non-empty)`, or for CSV `Wrote 98,765 bytes to /path/to/audit.csv (1,234 events across 3 days, 2 non-empty)`.
 2. **Scope used** (`org` or `tenant`) and any `--tenant-id` override.
 3. **Time window** — explicit ISO bounds, even if they came from a relative phrase ("last 7 days").
 4. **Filters applied** — sources, types, users, status.

package/skills/uipath-admin/references/ip-restriction/ip-restriction-commands.md

@@ -289,7 +289,7 @@ uip admin ip-restriction bypass-rules delete <RULE_ID> --output json
 
 ### `my-ip`
 
-Return the public IP the platform sees for the current caller. Use before `enforcement enable` to verify the allowlist covers your IP.
+Return the public IP the platform sees for the current caller. This is the direct answer whenever a user asks what their public IP is or what IP the platform sees for them ("what's my IP?", "what public IP does the platform see for me right now?") — run it standalone, no enforcement context needed. It is also the safety pre-flight before `enforcement enable` (verify the allowlist covers your IP).
 
 ```bash
 uip admin ip-restriction my-ip --output json

package/skills/uipath-agents/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: uipath-agents
 description: "End-to-end work with UiPath Agents of all types: build, integrate with UiPath Products (e.g., Orchestrator, Flow, Maestro), design with UiPath Tools (e.g., Agent Builder/Studio Web), and deploy. Covers Coded Agents (e.g., LangGraph, LlamaIndex, OpenAI Agents) and Low-Code Agents (`agent.json` / Agent Builder). Excludes coded functions / Functions SDK (separate skill)."
-when_to_use: "Must use when user mentions or implies any Agent lifecycle phase - e.g., auth, design, scaffold, Studio Web sync, flow integration, editing, pack/deploy/version bump, eval, tracing, guardrails, memory spaces, bindings, attachments. Example requests: 'create/build a UiPath agent', 'build a low-code / Agent Builder agent', 'add agent memory spaces', 'build a coded / Python agent (LangGraph / LlamaIndex / OpenAI Agents)', 'scaffold an agent project', 'run / evaluate / deploy my agent'."
+when_to_use: "Must use when user mentions or implies any Agent lifecycle phase - e.g., auth, design, scaffold, Studio Web sync, flow integration, editing, pack/deploy/version bump, eval, debug, tracing, guardrails, memory spaces, bindings, attachments. Example requests: 'create/build a UiPath agent', 'build a low-code / Agent Builder agent', 'add agent memory spaces', 'build a coded / Python agent (LangGraph / LlamaIndex / OpenAI Agents)', 'scaffold an agent project', 'run / debug / evaluate / deploy my agent'."
 allowed-tools: Bash, Read, Write, Edit, Glob, Grep, AskUserQuestion, WebFetch
 user-invocable: true
 ---
@@ -46,15 +46,16 @@ Determine the agent mode before proceeding:
 | Create or scaffold a new low-code agent project | Low-code | [lowcode/lowcode.md](references/lowcode/lowcode.md) § Quick Start | `lowcode/project-lifecycle.md`, `lowcode/agent-definition.md` |
 | Edit `agent.json` (prompts, model, schemas, contentTokens, entry-points.json) | Low-code | [lowcode/lowcode.md](references/lowcode/lowcode.md) § Capability Registry | `lowcode/agent-definition.md` |
 | Add a low-code tool (Orchestrator process — RPA / agent / API / agentic — or Integration Service) | Low-code | [lowcode/lowcode.md](references/lowcode/lowcode.md) § Capability Registry | `lowcode/capabilities/process/*`, `lowcode/capabilities/integration-service/*` |
+| Add an MCP (Model Context Protocol) server tool to a low-code agent | Low-code | [lowcode/lowcode.md](references/lowcode/lowcode.md) § Capability Registry | `lowcode/capabilities/mcp/mcp.md` |
 | Accept files as agent input / call the Analyze Files built-in tool | Low-code | [lowcode/capabilities/built-in-tools/built-in-tools.md](references/lowcode/capabilities/built-in-tools/built-in-tools.md) | `lowcode/capabilities/built-in-tools/analyze-attachments.md`, `lowcode/agent-definition.md` § File Attachments |
 | Work with file attachments (input, output, or created mid-run) | Coded | [coded/capabilities/file-attachments.md](references/coded/capabilities/file-attachments.md) | |
 | Build a conversational (chat-style) coded agent | Coded | [coded/capabilities/conversational-agents.md](references/coded/capabilities/conversational-agents.md) | |
 | Add a low-code context (Context Grounding RAG / attachments / DataFabric entity set) | Low-code | [lowcode/lowcode.md](references/lowcode/lowcode.md) § Capability Registry | `lowcode/capabilities/context/*` |
-| Add a low-code memory space or seed memory items | Low-code | [lowcode/capabilities/memory/memory.md](references/lowcode/capabilities/memory/memory.md) | use `uip agent memory`, then validate/migrate |
+| Add a low-code memory space or seed memory items | Low-code | [lowcode/capabilities/memory/memory.md](references/lowcode/capabilities/memory/memory.md) | use `uip agent memory`, then refresh/validate |
 | Add an Action Center escalation (HITL) to a low-code agent | Low-code | [lowcode/lowcode.md](references/lowcode/lowcode.md) § Capability Registry | `lowcode/capabilities/escalation/escalation.md` |
 | Add guardrails (PII, harmful content, custom rules) to a low-code agent | Low-code | [lowcode/lowcode.md](references/lowcode/lowcode.md) § Capability Registry | `lowcode/capabilities/guardrails/guardrails.md` |
 | Add guardrails to a coded agent | Coded | [coded/capabilities/guardrails/guardrails.md](references/coded/capabilities/guardrails/guardrails.md) | fetch official docs via WebFetch, ask middleware vs decorator, read agent code, write Python |
-| Add escalation guardrail (escalate action / Action Center app) | Low-code | [lowcode/capabilities/guardrails/guardrails.md](references/lowcode/capabilities/guardrails/guardrails.md) § escalate — Hand Off to Action Center | Run `uip solution resource list --kind App` to confirm app exists |
+| Add escalation guardrail (escalate action / Action Center app) | Low-code | [lowcode/capabilities/guardrails/guardrails.md](references/lowcode/capabilities/guardrails/guardrails.md) § escalate — Hand Off to Action Center | Run `uip solution resources list --kind App --source remote --output json` to confirm app exists |
 | Recommend guardrails for a low-code agent based on its context | Low-code | [lowcode/capabilities/guardrails/guardrails-recommend.md](references/lowcode/capabilities/guardrails/guardrails-recommend.md) | fetch catalog + list, analyze agent context, apply + validate |
 | Recommend guardrails for a specific scope or tool | Low-code | [lowcode/capabilities/guardrails/guardrails-recommend.md](references/lowcode/capabilities/guardrails/guardrails-recommend.md) § Scoped or Tool-Specific Filtering | filter candidates by scope or tool name after catalog analysis |
 | Validate whether existing guardrails are correctly configured | Low-code | [lowcode/capabilities/guardrails/guardrails-recommend.md](references/lowcode/capabilities/guardrails/guardrails-recommend.md) § Validate Mode | check correctness, actionability, and relevance per guardrail |
@@ -65,6 +66,7 @@ Determine the agent mode before proceeding:
 | Run low-code evaluations | Low-code | [lowcode/evaluations/evaluate.md](references/lowcode/evaluations/evaluate.md) | `lowcode/evaluations/evaluators.md`, `lowcode/evaluations/evaluation-sets.md`, `lowcode/evaluations/running-evaluations.md` |
 | Run offline evals for a published Orchestrator package | Low-code | [lowcode/evaluations/orchestrator-eval-run.md](references/lowcode/evaluations/orchestrator-eval-run.md) | Use `uip or eval run-offline-evals` (requires package published to Orchestrator) |
 | Validate, pack, publish, upload, or deploy a low-code agent | Low-code | [lowcode/lowcode.md](references/lowcode/lowcode.md) | `lowcode/project-lifecycle.md`, `lowcode/solution-resources.md` |
+| Debug / run a low-code agent end-to-end and inspect its output | Low-code | [lowcode/debug.md](references/lowcode/debug.md) | `lowcode/project-lifecycle.md` § `uip agent debug` |
 | Embed coded agent in a flow (solution-level) | Coded | [coded/embedding-in-flows.md](references/coded/embedding-in-flows.md) | |
 | Use coded agent in a flow | Coded | [coded/flow-integration.md](references/coded/flow-integration.md) | |
 | Use coded agent as tool for another agent in flow | Coded | [coded/flow-integration.md](references/coded/flow-integration.md) § Pattern 3 | |

package/skills/uipath-agents/references/coded/embedding-in-flows.md

@@ -95,3 +95,9 @@ uip maestro flow registry get "uipath.core.agent.<resourceKey>" --local --output
 The second command's `Data.Node` object is what the flow skill pastes into the flow's `definitions[]`. For the node instance shape and top-level `bindings[]` entries, see [agent/impl.md § In-solution variant](../../../uipath-maestro-flow/references/plugins/agent/impl.md#node-instance-inside-nodes--in-solution-variant).
 
 Without `--local`, `registry list`/`get` query the tenant registry (Orchestrator-published resources only) and will not surface the sibling project.
+
+## Wiring the Agent's Inputs
+
+For inputs that reference flow variables, use `{ "type": "literal", "expression": "{{ $vars.X }}", "fieldType": "string" }` — NOT `=js:...` expressions. `=js:` ships as a literal string to the agent activity and fails at runtime with `Cannot find name '<identifier>'`.
+
+Input-only rule. Mapping the agent's output back to a flow-level global on an End node DOES use `=js:` — see [variables-and-expressions.md § Variable Updates](../../../uipath-maestro-flow/references/shared/variables-and-expressions.md#variable-updates-variableupdates).

package/skills/uipath-agents/references/coded/flow-integration.md

@@ -9,6 +9,7 @@ Coded agents in flows use the `uipath.core.agent.{key}` node with `Orchestrator.
 The coded agent lives as a sibling folder inside the same solution as the flow. The flow references it with `section: "In this solution"`; the runtime resolves the node via the Studio Web projects API.
 
 - **Agent-side scaffolding:** [embedding-in-flows.md](embedding-in-flows.md)
+- **Wiring the agent's inputs:** [embedding-in-flows.md § Wiring the Agent's Inputs](embedding-in-flows.md#wiring-the-agents-inputs)
 - **Flow node JSON shape + top-level `bindings[]` + `definitions[]` entry:** [agent/impl.md § In-solution variant](../../../uipath-maestro-flow/references/plugins/agent/impl.md#node-instance-inside-nodes--in-solution-variant)
 
 The node-type's `{key}` is the local `resource.key` minted by `uip solution project add` (written to `resources/solution_folder/process/agent/<name>.json`) and surfaced by `uip maestro flow registry list --local`.

package/skills/uipath-agents/references/coded/frameworks/langgraph-integration.md

@@ -143,7 +143,7 @@ llm = UiPathAzureChatOpenAI(
 )
 ```
 
-**Model names:** Pass the model identifier as a string. Use `sdk.agenthub.get_available_llm_models()` to list the models available in your tenant.
+**Model names:** Pass the model identifier as a string. Run `uip codedagent list-models` to list the models available in your tenant.
 
 **Features:**
 - No API key needed — uses UiPath authentication
@@ -214,7 +214,7 @@ result: Analysis = Analysis.model_validate(raw_dict)
 | Multi-vendor flexibility (OpenAI, Anthropic, Google) | `UiPathChat` |
 | Specialized domains (e.g., code) | `UiPathChat` with a Claude model |
 
-Call `sdk.agenthub.get_available_llm_models()` to see the model strings available in your tenant and pass the exact `model_name` to the class constructor.
+Run `uip codedagent list-models` to see the model strings available in your tenant and pass the exact `model_name` to the class constructor.
 
 ---
 

package/skills/uipath-agents/references/coded/frameworks/llamaindex-integration.md

@@ -106,7 +106,7 @@ from uipath_llamaindex.llms import UiPathOpenAI, OpenAIModel
 llm = UiPathOpenAI(model=OpenAIModel.GPT_4O_2024_11_20)
 ```
 
-The enum is `OpenAIModel` (singular). `GeminiModel` and `BedrockModel` are also available from `uipath_llamaindex.llms` for non-OpenAI vendors. Use `sdk.agenthub.get_available_llm_models()` to list the models available in your tenant.
+The enum is `OpenAIModel` (singular). `GeminiModel` and `BedrockModel` are also available from `uipath_llamaindex.llms` for non-OpenAI vendors. Run `uip codedagent list-models` to list the models available in your tenant.
 
 ---
 

package/skills/uipath-agents/references/coded/frameworks/openai-agents-integration.md

@@ -4,6 +4,8 @@ A lightweight integration for running OpenAI Agents SDK agents on the UiPath pla
 
 > **Note:** This is a streamlined integration focused on agent execution and streaming. It does **not** support Human-in-the-Loop (HITL), state persistence/checkpointing, or UiPath process invocation from within agents. For workflows that need those features, use the [LangGraph](langgraph-integration.md) or [LlamaIndex](llamaindex-integration.md) integrations instead.
 
+> **STOP — input shape decides framework fit.** OpenAI Agents accept **only** a `messages` conversation input ([Input](#input) § CONSTRAINT). When the requested input is not a conversation, do NOT map it onto `messages` and proceed — STOP before scaffolding, tell the user, and let them choose another framework (e.g. [LangGraph](langgraph-integration.md)).
+
 ## Scaffolding a New Project
 
 If there is **no existing agent code**, scaffold an OpenAI Agents project with:
@@ -261,7 +263,9 @@ Then in `openai_agents.json`:
 
 ### Input
 
-All OpenAI Agents accept a `messages` field (for the LLM conversation) **plus** context fields from the Pydantic model:
+> **CONSTRAINT — `messages` is mandatory and cannot be removed.** The OpenAI Agents runner only accepts conversation input: a string (treated as a user message) or a list of message objects in the OpenAI Responses API format ([SDK docs — The agent loop](https://openai.github.io/openai-agents-python/running_agents/)). UiPath surfaces this as the `messages` field, so `uip codedagent init` ALWAYS emits `messages` as a **required** property in the entry-point input schema — every OpenAI Agents agent carries it, with or without a context model. Context fields from `Agent[ContextType]` are added **alongside** `messages`, never instead of it. There is no way to produce an OpenAI Agents input contract without `messages`. If the agent must accept a strict structured input with no `messages` field, OpenAI Agents cannot express it — pick a different framework (see [build.md](../lifecycle/build.md) § framework selection).
+
+OpenAI Agents accept the required `messages` field (the LLM conversation) **plus** optional context fields from the Pydantic model:
 
 ```json
 {"messages": "What is the weather?", "location": "New York"}

package/skills/uipath-agents/references/coded/lifecycle/build.md

@@ -46,6 +46,7 @@ Load capability references **only if the task requires them** — do not preload
 ## Additional Instructions
 
 - **Select a framework before writing any code.** Infer from the prompt if possible (tools/orchestration → LangGraph, RAG → LlamaIndex, simple LLM → OpenAI Agents, no LLM → Coded Function). If ambiguous, ask the user to choose.
+- **Structured input contract → not OpenAI Agents.** OpenAI Agents always require a `messages` input field and cannot express an input contract without it (see `../frameworks/openai-agents-integration.md` § Input). When the user needs a strict typed/structured input (e.g. a single named field, no `messages`), choose LangGraph (custom `StateGraph` with arbitrary input state) instead. Do NOT silently fall back to a Coded Function to satisfy the input shape — a Coded Function produces `ProjectType: Function`, not a coded agent, so it does not fulfill a request for an agent.
 - **Read ONLY the single framework reference** for the selected framework before writing code. Do NOT read other framework references or capability references unless the task explicitly requires that capability.
 - **Clean generated scaffold code before schema init.** After `uip codedagent new` and before running `uip codedagent init`, inspect `main.py` and remove scaffold hazards: no module-level `UiPathChat`, `UiPathAzureChatOpenAI`, `UiPath`, or other auth-dependent clients; instantiate LLM/SDK clients inside graph nodes/functions only; ensure importing `main.py` works without UiPath auth.
 - **NEVER instantiate LLM clients or SDK clients at module level.** `uip codedagent init` imports your Python file to introspect schemas — module-level `UiPathAzureChatOpenAI()`, `UiPathChat()`, `UiPathChatOpenAI()`, or `UiPath()` will fail because auth may not have happened yet. Always create these instances inside functions or graph nodes, never at the top level of the module.

package/skills/uipath-agents/references/coded/lifecycle/evaluate.md

@@ -40,23 +40,25 @@ evaluations/
 ├── eval-sets/
 │   └── smoke-test.json              # Test cases
 └── evaluators/
-    └── llm-judge-trajectory.json    # Evaluator config
+    └── llm-judge-output.json        # Evaluator config
 ```
 
 Every evaluator referenced in an eval set's `evaluatorRefs` must have a matching config file in `evaluations/evaluators/` — the `id` in the config must match the `evaluatorRefs` value exactly. Evaluators are auto-discovered from this directory.
 
-Example `evaluations/evaluators/llm-judge-trajectory.json`:
+Pick by output type: deterministic/structured → `uipath-exact-match` / `uipath-contains` / `uipath-json-similarity`; natural language → `uipath-llm-judge-output-semantic-similarity` (shown below). Use trajectory/tool-call evaluators only for multi-step / tool-using agents — they score 0.0 on single-step agents. Full guide: [evaluators.md](evaluations/evaluators.md), [best-practices.md](evaluations/best-practices.md).
+
+Example `evaluations/evaluators/llm-judge-output.json`:
 
 ```json
 {
   "version": "1.0",
-  "id": "LLMJudgeTrajectoryEvaluator",
-  "evaluatorTypeId": "uipath-llm-judge-trajectory-similarity",
+  "id": "LLMJudgeOutputEvaluator",
+  "evaluatorTypeId": "uipath-llm-judge-output-semantic-similarity",
   "evaluatorConfig": {
-    "name": "LLMJudgeTrajectoryEvaluator",
+    "name": "LLMJudgeOutputEvaluator",
     "model": "gpt-4o-mini-2024-07-18",
     "defaultEvaluationCriteria": {
-      "expectedAgentBehavior": "Agent should process the input and return a response."
+      "expectedOutput": {"<output_field>": "A correct, on-topic response for the given input."}
     }
   }
 }

package/skills/uipath-agents/references/coded/quickstart.md

@@ -118,40 +118,43 @@ Then STOP and wait. On reply, run the matching one-shot login from [../authentic
    Skip init only when the edit is purely inside node bodies / helpers (logic, prompts, business rules) and leaves every schema and the entry signature byte-identical. Then test locally with `uip codedagent run <ENTRYPOINT> '<input>'` (use the entrypoint name from `entry-points.json`, e.g., `main`).
 7. **Evaluate** — Run `uip codedagent eval <ENTRYPOINT> evaluations/eval-sets/smoke-test.json --no-report`. Idempotent by `has_evaluators` / `has_smoke_set`: create the missing one(s) only — **never overwrite an existing evaluator config or smoke set**, the user may have tuned them.
 
-   **Note:** `uipath-llm-judge-trajectory-similarity` requires emitted trace spans to populate `AgentRunHistory`. Plain `StateGraph` agents without explicit OpenTelemetry tracing produce empty history → all cases score 0.0 even on successful runs. If the smoke set scores 0.0, verify the agent actually executed (via local `uip codedagent run`) before treating it as a logic failure; consider an output-only evaluator for non-conversational graphs.
+   **Default the smoke evaluator to an output-based type, never a trajectory or tool-call evaluator** (those score 0.0 on single-step agents — use them only for multi-step / tool-using agents). Pick:
 
-   **If `has_evaluators == false`**, create `evaluations/evaluators/llm-judge-trajectory.json`. If the default `model` below is not available in the user's tenant, call `sdk.agenthub.get_available_llm_models()` and substitute a `model_name` from the returned list.
+   - **Deterministic or structured output** (a fixed string, number, or JSON shape) → `uipath-exact-match`, `uipath-contains`, or `uipath-json-similarity`. No LLM, no tenant model needed, binary/continuous scoring. Prefer this whenever the task allows it.
+   - **Natural-language output** (summaries, reports, free text) → `uipath-llm-judge-output-semantic-similarity` (`LLMJudgeOutputEvaluator`). Scores output semantics, works without tracing.
+
+   **If `has_evaluators == false`**, create `evaluations/evaluators/llm-judge-output.json` (default for NL output; swap to a deterministic type above when the output is fixed/structured). For any `uipath-llm-judge-*` type, if the default `model` below is not available in the user's tenant, run `uip codedagent list-models` and substitute an available model name.
 
    ```json
    {
      "version": "1.0",
-     "id": "LLMJudgeTrajectoryEvaluator",
-     "evaluatorTypeId": "uipath-llm-judge-trajectory-similarity",
+     "id": "LLMJudgeOutputEvaluator",
+     "evaluatorTypeId": "uipath-llm-judge-output-semantic-similarity",
      "evaluatorConfig": {
-       "name": "LLMJudgeTrajectoryEvaluator",
+       "name": "LLMJudgeOutputEvaluator",
        "model": "gpt-4o-mini-2024-07-18",
        "defaultEvaluationCriteria": {
-         "expectedAgentBehavior": "Agent should process the input and return a response."
+         "expectedOutput": {"<output_field>": "A correct, on-topic response for the given input."}
        }
      }
    }
    ```
 
-   **If `has_smoke_set == false`**, create `evaluations/eval-sets/smoke-test.json` with 2-3 test cases based on the agent's input schema (version is string `"1.0"`, top-level `id`/`name` required, test cases in `evaluations` array):
+   **If `has_smoke_set == false`**, create `evaluations/eval-sets/smoke-test.json` with 2-3 test cases based on the agent's input/output schema (version is string `"1.0"`, top-level `id`/`name` required, test cases in `evaluations` array). Key each case's criteria on the evaluator `id` you created above, and shape `expectedOutput` to match the agent's actual output field(s):
    ```json
    {
      "version": "1.0",
      "id": "smoke-test",
      "name": "Smoke Test",
-     "evaluatorRefs": ["LLMJudgeTrajectoryEvaluator"],
+     "evaluatorRefs": ["LLMJudgeOutputEvaluator"],
      "evaluations": [
        {
          "id": "test-1",
          "name": "Basic test",
-         "inputs": {"field": "value"},
+         "inputs": {"<input_field>": "value"},
          "evaluationCriterias": {
-           "LLMJudgeTrajectoryEvaluator": {
-             "expectedAgentBehavior": "Agent should process the input and return a response."
+           "LLMJudgeOutputEvaluator": {
+             "expectedOutput": {"<output_field>": "A correct, on-topic response for this input."}
            }
          }
        }
@@ -290,10 +293,12 @@ Execute the following in order, end-to-end, in one pass — do not pause for con
    The second command's `Data.Node` object is what gets pasted verbatim into the flow's top-level `definitions[]` array.
 
 7. **Wire the agent node into the `.flow` file.** Edit `<FlowName>.flow` directly:
-   - Add a `uipath.core.agent.<resourceKey>` node to `nodes[]` with `inputs.detail` matching the agent's input schema and `model.section: "In this solution"`.
+   - Add a `uipath.core.agent.<resourceKey>` node to `nodes[]` with one `inputs.<field>` entry per property in the agent's input schema (see step 6's `Data.Node.inputDefinition`) and `model.section: "In this solution"`.
+   - For input field values, see [embedding-in-flows.md § Wiring the Agent's Inputs](embedding-in-flows.md#wiring-the-agents-inputs).
    - Add the definition from step 6 to `definitions[]`.
    - Add a top-level `bindings[]` entry for the agent (no duplicates per `(resourceKey, propertyAttribute)`).
    - Add edges from upstream nodes to the agent's input port and from its output port downstream.
+   - To surface the agent's output as a flow-level result, declare an `out` global and map it on the End node with `"source": "=js:$vars.<agentNodeId>.output.<field>"` (outputs DO use `=js:`; only inputs do not).
 
    See [embedding-in-flows.md](embedding-in-flows.md) for the directory layout and [flow-integration.md § Pattern 1](flow-integration.md#pattern-1-in-solution-coded-agent) for the JSON shape.