@uipath/delegate-sdk 1.203.2 → 1.203.3

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/dist/index.mjs CHANGED
@@ -1,4 +1,4 @@
1
1
  [diffend] Oversized file quarantined before diffing.
2
2
  name: package/dist/index.mjs
3
- size: 56448375 bytes
4
- sha256: adfb586a3be38750c7ec8c57c2114990b741a1a518ebf5ab248a187ba2eab954
3
+ size: 56461437 bytes
4
+ sha256: f8df0d71131d892b032686736a413149e5bd31e1b855e435342eee96d00d9645
@@ -172,6 +172,8 @@ Infer BPMN vs Case vs RPA from the user's description using the signals below
172
172
 
173
173
  A confirmed task locks the RPA route for the whole journey — no further routing question ever fires, and TO-BE opens directly in `to-be-design.md`. When it looks like a process, ask nothing and proceed.
174
174
 
175
+ **What the user says about the work outranks how it looked to you.** A task is one bounded sequence a single person runs the same way every time; a predictable sequence that hands off between teams is a process, however fixed its steps; anything the user describes as varying by case, branching on judgment, or tracked per item is not one, however mechanical the individual steps sound. Take the description over the impression, and drop the task reading without asking again. This decision locks the whole journey and fires no later question, so a wrong one here is the only routing mistake with no recovery point — and it is reached at the opener, on the least evidence you will ever have. When the description and the step detail disagree, the description wins.
176
+
175
177
  **AS-IS capture format — ask before drawing (existing work).** Today's process can itself be captured as a case, a BPMN model, or a simple flow. Three inputs decide it outright and skip the question: a user-supplied BPMN model (saved verbatim), a Decision 1-confirmed task (Simple flow map), and a format the user already named — in this session or a wiki-recorded choice from an earlier one. That list is exhaustive. Sources that merely *suggest* a shape do not decide it: reading stages and owners out of a document is what makes Case map the **recommendation**, not what makes the question unnecessary. In every other case ask explicitly as part of the pre-drawing batch, recommendation first:
176
178
 
177
179
  > "How should I map today's process?"
@@ -188,11 +190,16 @@ Those are the only options Decision 2 offers. Never add a Flow, orchestration, o
188
190
 
189
191
  Batch Decision 2 with the TO-BE intent questions in one AskUser. Skip either decision when the user already explicitly named the target themselves — RPA for Decision 1, Case or BPMN for Decision 2 (state the route in one line and proceed). A source document's wording never names the target: a stakeholder saying the work "should be a flow" or wants "a fully orchestrated flow" is a signal for the recommendation, not an answer to the question. The answer holds through TO-BE, the PDD, and the solution design; only a genuine later contradiction reopens it — including a confirmed "task" that grows into a multi-actor process mid-design, which reopens routing via the escape hatch rather than silently continuing. If the user explicitly asks for Case instead of BPMN, honor Case.
190
192
 
193
+ **How the user says the work moves decides Decision 2, over any earlier recommendation.** Work that is long-running, sits with different owners, carries approvals and SLAs, or loops back for rework is Case, even across systems and handoffs. A path that is predictable and runs the same way every time through handoffs between teams is BPMN, even when each request reaches its own outcome — every request-driven process has an item with an outcome, so naming one decides nothing. The AS-IS map format was recommended before the user said this; when their description points the other way, recommend the route it points to and say so in one line. Decision 2 has two answers: deterministic, automatable steps are what the automation inside the chosen route does, never a third route.
194
+
195
+ **A recurring run is BPMN.** A process that repeats on a calendar and tracks nothing between runs — a scheduled report, batch, reconciliation or submission cycle — is BPMN even when it has stages, owners and SLAs: inside a cycle those signals describe the run, not a case lifecycle.
196
+
191
197
  - BPMN signals (feed Decision 2's recommendation): cross-functional handoffs, swimlanes, departments, events, gateways, message/timer waits, subprocesses, process model language, "BPMN", "process diagram", "process orchestration", "flow" or "orchestrated flow" said of a fixed sequence every item follows, or a broad business process whose target artifact is the process model itself.
192
198
  - Case signals (feed Decision 2's recommendation): a long-running business object, stages, status changes, SLAs, approvals, exception stages, rework loops, multiple owners, evidence/artifacts, or user language like "case", "request", "application", "claim", "ticket", or "lifecycle".
193
199
  - RPA signals (feed Decision 1 — task detection): deterministic screen steps, clicks, data entry, repetitive task execution, attended/unattended bot language, a fixed start/end transaction, or user language like "automate these steps", "bot", "workflow", or "RPA".
194
200
  - Mixed signals: Decision 1 treats it as a task only when a deterministic execution path dominates; at Decision 2, recommend Case when the request is primarily about managing a lifecycle and its states, BPMN when the user primarily needs a business process model across participants, events, decisions, and handoffs.
195
201
  - BPMN-vs-Case checkpoint: when BPMN is likely but Case would also be a defensible industry pattern, Decision 2 presents both explicitly — "BPMN process model (Recommended)" when handoffs/events/gateways dominate and "Case design" when a request/object is tracked through stages, owners, SLAs, and terminal outcomes. Explain the tradeoff in business terms inside the question, then continue with the selected guide.
202
+ - **Name the evidence in the recommendation.** State the two or three captured facts it rests on, in business terms, inside the question. A recommendation with no named fact is not made — say the signals are mixed and present both options level instead. Recommending a route the captured facts do not support is the failure this whole section exists to prevent: every downstream artifact, guide and gate follows the route, so a wrong one is not recoverable later in the phase.
196
203
 
197
204
  Route with this matrix:
198
205
 
@@ -505,7 +512,7 @@ Its anchors still carry no `sources=`, so no section ever shows a drift badge
505
512
 
506
513
  On success, call `AskUserFeedback` with `checkpoint: 'ba.sddGenerated'` (one non-blocking call; do not announce it or wait for a rating; skip silently if unavailable), then tell the user the solution design is ready in the document panel and name any ⚠ flags or open review items it carries. When planner's handoff still reads `Status: draft`, say the design is ready to review and name its blocking items — answering them is what lets planner flip it to `ready`.
507
514
 
508
- This turn first delivers the SDD, so its second ```sf follow-up suggestion is `Publish this SDD to Automation Hub` — the first stays the flow's next step. Silence conditions: "Automation Hub publishing" below; a publish request → read `automation-hub-publishing.md` in full.
515
+ This turn first delivers the SDD, so its second ```sf follow-up suggestion is `Publish this SDD to a business process in Automation Hub` — the first stays the flow's next step. Silence conditions: "Automation Hub publishing" below; a publish request → read `automation-hub-publishing.md` in full.
509
516
 
510
517
  That first ```sf follow-up suggestion is `Refine this SDD`. Phase 4 ends with the delivered SDD: it never offers or starts a build.
511
518
 
@@ -538,7 +545,7 @@ Automation Hub is the system of record for an approved process. Publishing is op
538
545
 
539
546
  Conversation Economy governs this: the offer rides a ```sf follow-up chip, so it costs no checkpoint, no `AskUser`, and no prose.
540
547
 
541
- On the turn a process document is first delivered (PDD or SDD), emit two ```sf suggestions. The first stays the flow's next step. The second is the offer, in the user's own voice, naming the type just delivered — `Publish this PDD to Automation Hub` after a PDD, `Publish this SDD to Automation Hub` after an SDD. That second chip is the whole offer: a click arrives as a user request, and that request is itself the approval; a later request to publish carries the same approval, so act on it immediately.
548
+ On the turn a process document is first delivered (PDD or SDD), emit two ```sf suggestions. The first stays the flow's next step. The second is the offer, in the user's own voice, naming the type just delivered — `Publish this PDD to a business process in Automation Hub` after a PDD, `Publish this SDD to a business process in Automation Hub` after an SDD. That second chip is the whole offer: a click arrives as a user request, and that request is itself the approval; a later request to publish carries the same approval, so act on it immediately.
542
549
 
543
550
  Raise it on that turn only. Stay silent while:
544
551
 
@@ -134,6 +134,8 @@ One MD per entity, in the matching folder.
134
134
  | `entities/people/<role-or-name>.md` | any person, title, or team | name or role title, team/dept, involvement (initiates/approves/executes/receives/escalates), availability/timezone if relevant |
135
135
  | `entities/artifacts/<artifact>.md` | any file, export, report, spreadsheet, form | name + format (xlsx, pdf, csv, …), source system, producer/consumer, frequency |
136
136
 
137
+ **The entity name is the source's spelling, and every named system gets its own page.** A transcript or OCR'd document spells a product name inconsistently — `eWalk` in seven turns, `Ewok` in two is one real example. The most frequent spelling in the source is the page name and the name field; the variants go in the page as `Also transcribed as:` so a later read-back matches either. Never normalize a name to a form the source never used (`EWOK` for `eWalk`), and never pick the rarer variant. Two systems named in the same breath — "is it Salesforce, eWalk?" — are two entity pages, not a choice between them; capture both and record which is the system of record once that is known. This works the way the clock-unit rule below does: the document is read back against the source bundle, so a renamed system scores as a missing detail even when the process around it is right.
138
+
137
139
  For artifacts, when the user has shared a sample file, append a light schema (column names, key field, one sample row). No confirmation table. A field the process only carries through stays at that light schema.
138
140
 
139
141
  **A decision-driving data field gets a full row instead.** When a field decides a route, sets a binding, or satisfies a compliance check, record it with the six columns PDD §6.3 projects:
@@ -253,6 +255,8 @@ Maintain `as-is/data-inputs-outputs.md` as a running list — one subsection per
253
255
 
254
256
  When the user or a source states a regulatory, policy, or audit obligation (retention rules, approval mandates, regulated decisions, evidence requirements) → append to `as-is/compliance.md`: the obligation, its source (regulation / internal policy), where the process enforces it today, and the evidence retained. This page is the PDD §9 source — compliance facts parked anywhere else never reach the document.
255
257
 
258
+ **Citation status is part of the obligation.** A named instrument — circular, statute, directive, article number — whose text the sources do not contain is `unverified`. Record the name, what it is claimed to require, and the mark. Never cite its articles as authoritative and never derive specific requirements from it: a source may name an instrument that does not exist, and an uncorroborated citation otherwise reaches §9 as settled regulation. Say so to the user in one line and ask them to confirm it or route it to compliance — a "TBD: obtain the full text" next-step item does not discharge this, because the matrix above it still reads as verified. The mark travels to §9 with the row.
259
+
256
260
  ### Pain Points
257
261
 
258
262
  When the user expresses frustration, manual workarounds, errors, delays, or rework → append to `as-is/pain-points.md`:
@@ -161,6 +161,15 @@ The renderer parses with Mermaid, converts to a semantic graph, and classifies.
161
161
  | `conflicting-semantics` | two modes on one node, two case task types, kind contradicting the shape |
162
162
  | `source-too-large` | fenced source over 10,000 characters — the map holds step-level detail; collapse to sub-process nodes (Size failures below) |
163
163
 
164
+ **Keep a map under about 30 nodes.** The lint reports the node count but caps only
165
+ the source's length (`source-too-large`), so a map can grow past what a reader can
166
+ follow before it fails — and a map drawn wide enough to need it
167
+ costs the turns the document still needs. Above roughly 30,
168
+ collapse detail into summary nodes and let the per-stage or per-step pages carry
169
+ it: the pages are the design, the map is how it is read. A current-state map of a
170
+ large process is the usual case, and it is also the one where the future map has
171
+ not been drawn yet, so the budget spent here is taken from Phase 2 and 3.
172
+
164
173
  ## Diagram source audit
165
174
 
166
175
  After every write of a canonical `diagram.md`, **run the bundled lint against the saved file as the very next tool call** — never self-assess the source by eye and never stop at a successful WriteFile/EditFile result. Select the command from `PlatformContextType` `os.name`; the script directory is pre-resolved, substitute only the diagram path:
@@ -101,7 +101,7 @@ For each section: read its mapped wiki sources, then emit the section block exac
101
101
  For route-specific source-map cells, choose the path list for the confirmed route. Do not serialize route labels or paths from other routes.
102
102
  2. The section heading (H1 or H2 per template).
103
103
  3. Body:
104
- - Preserve source facts, table structure, numbers, and identifiers exactly; translate prose into the document language when the source uses another language.
104
+ - Preserve source facts, table structure, numbers, and identifiers exactly; translate prose into the document language when the source uses another language. Keep the named items a wiki page lists — inputs, checks, documents, criteria, policies — as named; never fold them into a summary label such as "commercial profile".
105
105
  - Follow the template's structure for this section + the shared content rules.
106
106
  - For `current-state` on New + RPA, New + Case, or New + BPMN, follow the template's new-work Current State structure. Do not invent AS-IS flow, metrics, pain points, or an AS-IS diagram.
107
107
  - For `current-state` on existing work, the step pages pick the §3.2 layout: `Recording:` ranges → the template's recording layout (with its §5 and §8.1 columns and the extra `sources=` paths); none → the default layout. The recording layout has no row cap: one row per step page.
@@ -183,7 +183,7 @@ A declared source that does not exist yet — a P0-gap section's intended path
183
183
 
184
184
  Run this phase's one self-review here — after assembly, before the draft is surfaced, and never again afterwards (SKILL.md "One Review Per Phase"): one pass, one fix round, and whatever it still leaves stays in the draft as an open item, named in the line below. A second review pass over the assembled draft is forbidden. A fact it corrects on a `verified` wiki page is named in the same reply that writes it (SKILL.md "Editing a confirmed page").
185
185
 
186
- Surface to the user: "Draft ready. Open the PDD panel to review it." — and when the draft carries open items, add one line naming how many and where they cluster (e.g. "3 open items flagged, mostly around compliance evidence"). Do not paste the draft into chat. Then call `AskUserFeedback` with `checkpoint: 'ba.pddGenerated'` (one non-blocking call; do not announce it or wait for a rating; skip silently if the tool is unavailable). This turn first delivers the PDD, so its second ```sf follow-up suggestion is `Publish this PDD to Automation Hub` — the first stays the flow's next step. Silence conditions: "Automation Hub publishing" in `SKILL.md`; a publish request → read `automation-hub-publishing.md` in full.
186
+ Surface to the user: "Draft ready. Open the PDD panel to review it." — and when the draft carries open items, add one line naming how many and where they cluster (e.g. "3 open items flagged, mostly around compliance evidence"). Do not paste the draft into chat. Then call `AskUserFeedback` with `checkpoint: 'ba.pddGenerated'` (one non-blocking call; do not announce it or wait for a rating; skip silently if the tool is unavailable). This turn first delivers the PDD, so its second ```sf follow-up suggestion is `Publish this PDD to a business process in Automation Hub` — the first stays the flow's next step. Silence conditions: "Automation Hub publishing" in `SKILL.md`; a publish request → read `automation-hub-publishing.md` in full.
187
187
 
188
188
  ---
189
189
 
@@ -223,7 +223,7 @@ Four H2 subsections, each a table:
223
223
 
224
224
  ### Section 12: Glossary
225
225
 
226
- Heading: `"12. Glossary"`. Source: `entities/` subfolders + abbreviations from transcripts. Keep to terms actually used in this PDD.
226
+ Heading: `"12. Glossary"`. Source: `entities/` subfolders + abbreviations and domain terms from transcripts. Keep to terms actually used in this PDD. When the source is in another language, give each domain term it names (documents, checks, criteria, policies) as the English term followed by the source's own term in italics — e.g. "Credit behaviour (*comportamiento crediticio*)" — so a reader can trace the term back to the source.
227
227
 
228
228
  Table (caption required, columns: Term | Definition). Alphabetical, one row per term.
229
229
 
@@ -2,7 +2,7 @@
2
2
  name: media-tools
3
3
  friendly_name: Media Tools
4
4
  version: 1.0.0
5
- description: "Edits local audio/video with ffmpeg/ffprobe (from PATH, else the build bundled in Delegate): trims/cuts clips, converts format or codec, compresses to a target file size, extracts audio, extracts frames, builds thumbnails or a contact sheet, makes GIFs, merges/concatenates clips, changes speed, resizes/scales, and burns in subtitles. Use when a task mentions cut, clip, trim, shorten, convert, compress, shrink, reduce size, rip audio, grab a frame or screenshot from a video, gif, stitch, join, concat, slow-mo, timelapse, rescale, or hardcode/burn captions on a media file. Input is a local file/resource path or a URL (download it first). Not for transcription (use transcribe-media) or recorded-task conversion."
5
+ description: "Edits local audio/video with ffmpeg/ffprobe (from the Delegate bundle, else PATH): trims/cuts clips, converts format or codec, compresses to a target file size, extracts audio, extracts frames, builds thumbnails or a contact sheet, makes GIFs, merges/concatenates clips, changes speed, resizes/scales, and burns in subtitles. Use when a task mentions cut, clip, trim, shorten, convert, compress, shrink, reduce size, rip audio, grab a frame or screenshot from a video, gif, stitch, join, concat, slow-mo, timelapse, rescale, or hardcode/burn captions on a media file. Input is a local file/resource path or a URL (download it first). Not for transcription (use transcribe-media) or recorded-task conversion."
6
6
  icon: FaFilm
7
7
  color: F97316
8
8
  ---
@@ -13,7 +13,7 @@ Edits one media file with ffmpeg/ffprobe. Input: a local file/resource path, or
13
13
 
14
14
  ## Resolve the binaries
15
15
 
16
- First action, before any recipe: read `references/resolve-binaries.md`, run its snippets, set `$FFMPEG`/`$FFPROBE`, and confirm `"$FFMPEG" -version` exits 0. Order, first hit wins: PATH → Delegate bundle → existing install → install (last, needs user consent).
16
+ First action, before any recipe: read `references/resolve-binaries.md`, run its snippets, set `$FFMPEG`/`$FFPROBE`, and confirm `"$FFMPEG" -version` exits 0. Order, first hit wins: Delegate bundle → PATH → existing install → install (last, needs user consent).
17
17
 
18
18
  ## Run a recipe
19
19
 
@@ -2,35 +2,40 @@
2
2
 
3
3
  Set `$FFMPEG` + `$FFPROBE` (Windows) or `FFMPEG` + `FFPROBE` (macOS) once, then reuse. Try strategies in order; first hit wins. `<skill-path>` = this skill's folder, from the enclosing `<skills_folder path>`.
4
4
 
5
- Order: **PATH → bundle → machine → install**. Stop at the first that resolves. Install modifies the machine — last resort, needs user consent.
5
+ Order: **bundle → PATH → machine → install**. Stop at the first that resolves. Install modifies the machine — last resort, needs user consent.
6
6
 
7
- Verify any candidate before using it: `"$FFMPEG" -version` (exit 0). Derive ffprobe next to ffmpeg if not separately found.
7
+ Verify any candidate before using it: `"$FFMPEG" -version` (exit 0). A bundle or PATH hit counts only with ffprobe beside it or found separately; otherwise try the next strategy. Derive ffprobe next to ffmpeg if not separately found.
8
8
 
9
- ## 1. PATH
9
+ ## 1. Bundle (Delegate interop)
10
10
 
11
- Fastest. ffmpeg installed and on PATH.
11
+ Delegate ships an ffmpeg build with its interop layer near this skill, or uses Robot's, named by `$env:UIPATH_FFMPEG_DIR`. On Windows try that folder, then the installed path `<skill-path>\..\..\interop\ffmpeg.exe`: a restricted shell may run them but not list folders. Otherwise search recursively three levels above the skill folder, which stays inside the Delegate install.
12
12
 
13
13
  ```powershell
14
- $FFMPEG = (Get-Command ffmpeg.exe -ErrorAction SilentlyContinue).Source
15
- $FFPROBE = (Get-Command ffprobe.exe -ErrorAction SilentlyContinue).Source
14
+ $S = "<skill-path>"
15
+ $FFMPEG = @($env:UIPATH_FFMPEG_DIR, (Join-Path $S "..\..\interop")) | Where-Object { $_ } |
16
+ ForEach-Object { Join-Path $_ "ffmpeg.exe" } | Where-Object { Test-Path -LiteralPath $_ -PathType Leaf } |
17
+ Select-Object -First 1
18
+ if (-not $FFMPEG) {
19
+ $FFMPEG = Get-ChildItem -Path (Join-Path $S "..\..\..") -Recurse -File -Filter ffmpeg.exe -ErrorAction SilentlyContinue |
20
+ Select-Object -First 1 -ExpandProperty FullName
21
+ }
16
22
  ```
17
23
  ```bash
18
- FFMPEG="$(command -v ffmpeg)"
19
- FFPROBE="$(command -v ffprobe)"
24
+ S="<skill-path>"
25
+ FFMPEG="$(find "$S/../../.." -type f -name ffmpeg -perm -u+x 2>/dev/null | head -n1)"
20
26
  ```
21
27
 
22
- ## 2. Bundle (Delegate interop)
28
+ ## 2. PATH
23
29
 
24
- Delegate ships an ffmpeg build with its interop layer near this skill. Subpath varies — don't assume one; search recursively under an ancestor of the skill folder.
30
+ ffmpeg on PATH: the bundle, which Delegate shells put first, or a user install.
25
31
 
26
32
  ```powershell
27
- $S = "<skill-path>"
28
- $FFMPEG = Get-ChildItem -Path (Join-Path $S "..\..\..\..") -Recurse -File -Filter ffmpeg.exe -ErrorAction SilentlyContinue |
29
- Select-Object -First 1 -ExpandProperty FullName
33
+ $FFMPEG = (Get-Command ffmpeg.exe -ErrorAction SilentlyContinue).Source
34
+ $FFPROBE = (Get-Command ffprobe.exe -ErrorAction SilentlyContinue).Source
30
35
  ```
31
36
  ```bash
32
- S="<skill-path>"
33
- FFMPEG="$(find "$S/../../../.." -type f -name ffmpeg -perm -u+x 2>/dev/null | head -n1)"
37
+ FFMPEG="$(command -v ffmpeg)"
38
+ FFPROBE="$(command -v ffprobe)"
34
39
  ```
35
40
 
36
41
  ## 3. Resolve on machine
@@ -70,12 +75,14 @@ brew install ffmpeg # macOS
70
75
  After a hit, normalize and derive ffprobe:
71
76
 
72
77
  ```powershell
73
- if (-not $FFMPEG) { throw "ffmpeg not resolved (PATH/bundle/machine/install all failed)" }
78
+ if (-not $FFMPEG) { throw "ffmpeg not resolved (bundle/PATH/machine/install all failed)" }
74
79
  $FFMPEG = (Resolve-Path $FFMPEG).Path
75
80
  if (-not $FFPROBE) { $FFPROBE = $FFMPEG -replace 'ffmpeg\.exe$', 'ffprobe.exe' }
76
81
  ```
77
82
  ```bash
78
- [ -n "$FFMPEG" ] || { echo "ffmpeg not resolved (PATH/bundle/machine/install all failed)" >&2; exit 1; }
83
+ [ -n "$FFMPEG" ] || { echo "ffmpeg not resolved (bundle/PATH/machine/install all failed)" >&2; exit 1; }
79
84
  FFMPEG="$(cd "$(dirname "$FFMPEG")" && pwd)/$(basename "$FFMPEG")"
80
85
  [ -n "$FFPROBE" ] || FFPROBE="$(dirname "$FFMPEG")/ffprobe"
81
86
  ```
87
+
88
+ No working ffmpeg + ffprobe (all strategies failed, or install declined) → tell the user in one plain line that ffmpeg isn't available. Don't retry shell calls.
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@uipath/delegate-sdk",
3
- "version": "1.203.2",
3
+ "version": "1.203.3",
4
4
  "description": "UiPath Delegate SDK - Programmatic agent interface",
5
5
  "license": "Apache-2.0",
6
6
  "type": "commonjs",
@@ -50,10 +50,10 @@
50
50
  "zod-to-json-schema": "^3.25.1"
51
51
  },
52
52
  "optionalDependencies": {
53
- "@uipath/delegate-runtime-darwin-arm64": "1.203.2",
54
- "@uipath/delegate-runtime-win32-x64": "1.203.2",
55
- "@uipath/delegate-runtime-win32-x64-driver": "1.203.2",
56
- "@uipath/delegate-runtime-linux-x64": "1.203.2"
53
+ "@uipath/delegate-runtime-darwin-arm64": "1.203.3",
54
+ "@uipath/delegate-runtime-win32-x64": "1.203.3",
55
+ "@uipath/delegate-runtime-win32-x64-driver": "1.203.3",
56
+ "@uipath/delegate-runtime-linux-x64": "1.203.3"
57
57
  },
58
58
  "devDependencies": {
59
59
  "@types/node": "^22.0.0",