Skip to content

fix(uipath-maestro-case): add test-recipient and go-live guidance [CA-232] - #4870

Merged
roberts-cliff merged 12 commits into
mainfrom
fix/case-test-recipient-guidance
Nov 18, 2026
Merged

roberts-cliff merged 12 commits into
mainfrom
fix/case-test-recipient-guidance

Conversation

@roberts-cliff

@roberts-cliff roberts-cliff commented Nov 17, 2026 •

Copy link
Copy Markdown
Contributor

Summary

  • The case skill had no guidance on action recipients or the go-live trap. recipient: '<email>' bakes the address into the plan. To go live, someone must edit every action and publish again.
  • PACT golden thread step 27 caught this. Autopilot routed actions to the user's own email and did not say what going live requires (CA-232).
  • New rules in case-runtime.md § Action recipients and the go-live trap:
    • RULE:case.recipient.externalize gives the recipe for a test or placeholder assignee: declare a case input with a test default, then set recipient: { type: 3, value: '=vars.<input>' }.
      • To go live, change the input default or pass the real address at case start. The actions stay unchanged.
      • The rule also lists the recipient types (0 user id, 1 group, 2 email, 3 expression). It says a bare '=vars.x' string is type 2 and is looked up as an email.
    • RULE:case.recipient.no-silent-default makes the final summary list every assignee that the user did not name as production: input defaults, literal recipients, toUser/toGroup escalation targets, and <UNRESOLVED: ...>. For each, the summary says what going live changes.
      • The rule gates the two commands that start a case instance, case debug and case process run. An <UNRESOLVED> assignee never runs.
      • An unconfirmed address needs the user's answer. A request or pre-approval that names the step counts as that answer.
      • upload, publish, check, compile and validate are not gated.

Evidence for type 3

  • CLI packages/case-tool/src/services/sdd-convert/recipient.ts maps an expression recipient to Type 3, the engine's Unknown (AppTasksFacade.ResolveAssignmentCriteria). The engine evaluates the value and assigns it as a user when it is an email, or as a group name otherwise. The same source comment says 68 of the 74 expression recipients in the eval corpus carry type 3 (tests/fixtures/strict/README.md matches), and that a type 3 recipient ran in demo TIF-12003.
  • CLI strict-case-rules.spec.ts accepts { Type: 3, Value: '=vars.x' }. It errors on =vars.x under types 0, 1 or 2.
  • Probe with SDK 6.120.17:
    • case-check --source on the recipe prints check: OK.
    • case-compile emits {"Type": 3, "Value": "=vars.approver"}, and the input default lands in variables.inputs[].default.
    • The same expression written as a bare string makes check warn STRICT_RECIPIENT_EXPRESSION_TYPE.
  • Not run:
    • uip maestro case validate --strict: the local uip cannot install its maestro-tool (npm edgesOut error).
    • case debug.
  • The SDK's own RecipientType doc comment still says 0/1/3 are "reserved". That comment is stale against the CLI rule and its own ported lint. It should be fixed in builder-sdk.

Not covered

CA-232 also asks for test settings that live outside the case, such as an Orchestrator asset. This PR moves the assignee into a case input, which the caller can override at start (uip maestro case process run --inputs). It does not read the assignee from an asset. That part of CA-232 stays open.

Earlier drafts

The first draft's snippet did not compile (types is not exported from /case) and used the Flow path vars.start.output.*. A later draft dropped the recipe because runtime support looked unconfirmed. The evidence above confirms it.

Test plan

  • Adversarial review clean
  • skill-lint CI passes
  • case-runtime.md is linked from SKILL.md's reference navigation
  • pytest tests/tasks/uipath-maestro-case/ -q: 459 passed, 2 skipped, 11 subtests passed
  • Smoke skill-case-edit-add-task-run-once-default (7/8 on an earlier head) is investigated separately

🤖 Generated with Claude Code

https://claude.ai/code/session_013tiDkBvodTad2Je4FVfz4q

@roberts-cliff
roberts-cliff requested review from a team, dmetzgar and tmatup as code owners November 17, 2026 18:55
@github-actions

github-actions Bot commented Nov 17, 2026 •

Copy link
Copy Markdown
Contributor

Claude finished @roberts-cliff's task in 1m 1s —— View job


Summary

Adds an "Action recipients and the go-live trap" section to skills/uipath-maestro-case/references/case-runtime.md with two rules: RULE:case.recipient.externalize and RULE:case.recipient.no-silent-default (CA-232). The goal is right. The TS snippet, though, doesn't match the SDK API the rest of this skill teaches, and agents copy snippets as written.

Change-by-Change Review

1. Snippet syntax — case-runtime.md:48-52

Severity: High

Both lines in the snippet disagree with the rest of this skill:

  • types.string — No file in the skill imports or uses types. The documented input shape is a string literal: examples/IntakeBinding.case.ts:28 has .input({ caseId: 'string' }, { from: manual }). If an agent copies the snippet, it gets an undefined identifier, or it tries to import types from the SDK.
  • '=vars.start.output.approver' — The skill says a case input is read as =vars.<name>. IntakeBinding.case.ts:11 says "readable as =vars.<name>", and case-runtime.md:78-79 says to declare every vars.x read with .input(). No vars.start.output.* path appears anywhere else. The reference should be '=vars.approver'.
  • The prose says "with a placeholder default", but the snippet never shows one, and no file in the skill shows how to give an .input() a default. An agent will either make up a syntax or skip the default. Either show the real default syntax or drop "with a placeholder default" from the prose.
  • The snippet leaves out { from: <trigger> }. If an unbound input is valid, that's fine. If the caller has to supply the address at case start, bind the input to the manual trigger, as IntakeBinding does.

Suggested snippet, assuming the SDK supports expressions in recipient:

.input({ approver: 'string' }, { from: manual })
// ...
.action({ ..., recipient: '=vars.approver' })

Fix this →

2. The rule's argument doesn't hold up — case-runtime.md:41-58

Severity: Medium

Line 41 says the trap is that a literal recipient "bakes the email address into the plan" and that going live means "editing … the source and republishing". But step 2 at lines 57-58 says the same thing about the input default: going live still means changing it in the source and publishing a new version. As written, an agent can't tell what externalizing gains. The real gains are:

  • one place to edit instead of every action, and
  • if there's no default, the caller supplies the address at runtime, so nothing needs republishing.

State these directly. Also say which to prefer: no default (caller supplies the address) for production, and a default only for test runs. Right now "a trap for go-live" is followed by a fix that still has the same trap.

3. Conflicts with SKILL.md and the examples

Severity: Medium

SKILL.md:92 (the quick-start an agent copies first) uses recipient: 'manager@corp.com' as a literal. So do examples/ClaimReviewSLA.case.ts:36,51, PermitIssuance.case.ts:31-49 and NotifyOnApproval.case.ts:30. With nothing in SKILL.md pointing to the new rule, the copied pattern is the exact one this rule forbids, and the rule sits in a reference the agent may never load. Do one of these:

  • add a one-line Critical Rule or a pointer in SKILL.md, or
  • add a comment in the quick-start that these literals are placeholders.

That puts the rule in front of the agent when it's writing the action.

4. RULE:case.recipient.no-silent-default — case-runtime.md:60-63

Severity: Low

The content is fine. It partly repeats "Tell the user before publishing" at lines 54-58; merge the two into one disclosure list. "Do not set…" is good prescriptive wording.

5. Flavor overrides

OK. The new section is outside every skill-flavor block. skill-flavors/studioweb/uipath-maestro-case/references/case-runtime.md only replaces existing named blocks, so studioweb picks up this section unchanged.

6. Content hygiene

OK. No secrets or personal paths. The only email-like strings are generic.

What's Missing

  • No coder_eval task covers this behaviour. A small tests/tasks/uipath-maestro-case/ task would do it: the prompt names a test approver's email, and the grading checks file_contains =vars. on the recipient and does not find a literal @ address there. This is the PACT regression (step 27), so it should have a regression test.
  • No pointer from SKILL.md to the new rule (see item 3).
  • No documented syntax for an .input() default (see item 1).

Area Ratings

Area Status Notes
Frontmatter OK Not touched
E2E Tests Issue No task for the CA-232 regression
Skill Body Issue SKILL.md quick-start contradicts the new rule
References & Assets Issue Wrong input and var-reference syntax in the snippet
Repo Hygiene OK One file, scoped to this skill

Issues for Manual Review

  • Whether .action({ recipient }) accepts an = expression at all, or only a literal email or toUser/toGroup scope. If it only takes literals, the whole externalize rule can't be followed. Check the generated API or compile a sample.
  • Whether .input() supports a default value, and with what syntax.
  • Whether an unbound .input() (no { from }) can be supplied by the caller at case start.

Conclusion

Request changes. The guidance is worth having. Before merging, fix the snippet (item 1) so agents don't learn invalid SDK syntax, and fix the reasoning (item 2) so the fix doesn't repeat the trap it describes. Items 3 and the test task are strongly recommended.

roberts-cliff and others added 2 commits November 18, 2026 09:04
…-232]

The case skill had no guidance on action recipients and the go-live
trap: a literal recipient email baked into the plan requires editing
and republishing to change. PACT golden thread caught it at step 27
when the agent set the user's email as a default without warning.

New RULE:case.recipient.externalize and RULE:case.recipient.no-silent-default
in case-runtime.md tell the agent to wire recipients through case
inputs and warn the user before publishing with a test default.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013tiDkBvodTad2Je4FVfz4q
The =vars. expression syntax on the recipient field is unverified by
any existing example or test. Add a verification step (case check)
and a fallback path in case the compiler treats it as a literal.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013tiDkBvodTad2Je4FVfz4q
@roberts-cliff
roberts-cliff force-pushed the fix/case-test-recipient-guidance branch from cd16754 to f1ba7da Compare November 18, 2026 16:04
tmatup
tmatup previously requested changes Nov 18, 2026

@tmatup tmatup left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Request changes: one blocking bug. The recipe in RULE:case.recipient.externalize doesn't build as written. After the obvious repair it emits a recipient that points at a variable the case doesn't have, and the step the rule says will catch that (case check) reports OK.

I compiled three variants with the SDK's own case-check / case-compile (builder-sdk main d8e9380), then ran uip maestro case validate on the output.

Variant case check compile case validate Emitted data.recipient
Doc example as written (types.string) crash: '@uipath/maestro-builder-sdk/case' does not provide an export named 'types' crash n/a n/a
Doc expression =vars.start.output.approver (with 'string') OK ok Valid {"Type": 2, "Value": "=vars.start.output.approver"}
Case convention =vars.approver OK ok Valid {"Type": 2, "Value": "=vars.approver"}

Meant vs. got:

meant:  input approver --> recipient resolves to the approver's email at runtime
got:    recipient = Type 2 ("a single user by email") with the literal string
        "=vars.start.output.approver"; the case exposes the input as vars.approver,
        and vars.start does not exist in a case plan

Why it can happen.

  • vars.start.output.* is the Flow v2 shape. Case inputs are read as =vars.<name> (see IntakeBinding.case.ts and the =vars. uses elsewhere in case-runtime.md).
  • The doc says to "verify with case check --source that the expression resolves". But checkCaseExpressions (builder-sdk typescript/sdk/src/case/case-expr-check.ts:189) scans gates, SLAs, task inputs and action form inputs. It never reads task.action.recipient. So the check cannot flag a wrong recipient, and the fallback ("if check reports the expression as a literal") never triggers.
  • I found no case plan in the builder-sdk platform corpus with an = expression in recipient.Value. Every one is a plain email. Runtime not verified: I don't know if Action Center evaluates an expression there at all, even =vars.approver.

Suggested fix.

  1. Run one action task with recipient: '=vars.approver' through case debug (or have the runtime owner confirm it) before telling agents to use it.
  2. If it works, fix the snippet to .input({ approver: 'string' }, { from: manual }) plus recipient: '=vars.approver', with the real default syntax the text refers to. The current snippet declares no default.
  3. Drop the claim that case check verifies the reference. Or, the enabling-layer fix: file a builder-sdk issue to add action.recipient to checkCaseExpressions (undeclared vars.X gives an error; a non-email literal gives a lint). Then the claim becomes true for every author, not only those who read this rule.

RULE:case.recipient.no-silent-default is fine as written.

Also: one smoke task is red on this head (skill-case-edit-add-task-run-once-default, 87.5% < 95%). It tests run-once defaults, not recipients. I did not root-cause it.

Not reviewed: runtime behavior of expression recipients, and SKILL.md routing to the new section.


Review run with Claude Code; each finding was reproduced with a probe at the PR head.

`case check --source` that the expression resolves:

```ts
.input({ approver: types.string })

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Blocking. types is not exported from @uipath/maestro-builder-sdk/case. Compiling this snippet fails with does not provide an export named 'types'. Case inputs use string type names plus their trigger: .input({ approver: 'string' }, { from: manual }). The text above promises a placeholder default, but this snippet declares none.

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fixed in a0e145e. The snippet now declares .input({ approver: { type: 'string', default: '<TEST_EMAIL>' } }, { from: manual }), which is the SDK's { type, default } input shape. With SDK 6.120.17, case-check --source prints check: OK on it, and case-compile puts the default in variables.inputs[].default.

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Verified at 5451ad2: the snippet now uses .input({ approver: { type: 'string', default: '<TEST_EMAIL>' } }, { from: manual }) and compiles with the builder SDK; the default lands in variables.inputs.

```ts
.input({ approver: types.string })
// ...
.action({ ..., recipient: '=vars.start.output.approver' })

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Blocking. =vars.start.output.approver is the Flow path shape. In a case plan the input is vars.approver (variables.inputOutputs[0].id). This compiles to {"Type": 2, "Value": "=vars.start.output.approver"} (Type 2 = a single user by email), and case check says OK and case validate says Valid. Runtime not verified: there is no expression-valued recipient anywhere in the platform corpus, so please confirm with one case debug that even =vars.approver resolves before recommending it.

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fixed in a0e145e. The recipient is now { type: 3, value: '=vars.approver' }, and it compiles to {"Type":3,"Value":"=vars.approver"}. Type 3 is the expression type: CLI sdd-convert/recipient.ts maps an expression recipient to it, and strict-case-rules.spec.ts accepts it. The same source comment says 68 of the 74 corpus expression recipients use type 3, and that one ran in TIF-12003. I did not run case debug.

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Verified at 5451ad2: recipient: { type: 3, value: '=vars.approver' } compiles to {"Type": 3, "Value": "=vars.approver"} and passes validate --strict. Type 3 is backed by the CLI's strict rule and its corpus note; runtime assignment not verified by me.

When a request names a person or test mailbox as the recipient, wire it
through a case **input** with a placeholder default instead of a literal.
Declare the input, reference it in the `recipient` field, and verify with
`case check --source` that the expression resolves:

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should fix: case check cannot verify this. checkCaseExpressions (builder-sdk case/case-expr-check.ts:189) never reads action.recipient. A wrong path passes as check: OK, so the fallback on line 56 never triggers. Either drop the verification claim, or file the builder-sdk change that adds recipient to the expression check.

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fixed in a0e145e. The verification claim now names what really catches this shape. With SDK 6.120.17, case check warns STRICT_RECIPIENT_EXPRESSION_TYPE on a bare-string '=vars.approver', because that is type 2. I probed this. case validate --strict fails on the same shape: strict-case-rules.ts raises it with severity error, which I read in the source but did not run. The script-task fallback is gone.

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Partly: the doc now says case check warns STRICT_RECIPIENT_EXPRESSION_TYPE, but that code is only in validate --strict. The SDK check says OK on the bare form. Details in the new inline comment; validate --strict does catch it, so it's not blocking.

@dmetzgar dmetzgar left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Agree with the blocking points in the existing review. @uipath/maestro-builder-sdk/case exports no types, so the snippet does not compile. =vars.start.output.approver is the Flow shape: a case input is =vars.<name>, and the wrong path compiles to a literal email recipient that case check does not catch, because checkCaseExpressions never reads recipient. Please fix the snippet, drop the case check --source verification claim, and confirm once with case debug that an expression-valued recipient resolves before recommending it.

roberts-cliff and others added 6 commits November 18, 2026 11:40
The snippet did not compile (types is not exported from /case), used the Flow
path shape vars.start.output.*, and case check never reads action.recipient.
No case plan in use carries an expression recipient. Keep the go-live warning;
tell agents to write a plain address and say what going live requires.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013tiDkBvodTad2Je4FVfz4q
…tate the recipient type

Say that a bare string compiles to type 2 (single user by email) and that
types 0/1/3 are unexercised. Drop the unproven validate claim. Fire the
disclosure for any recipient the user did not name as the production
assignee, and keep that text in one rule only.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013tiDkBvodTad2Je4FVfz4q
…ipients

Keep a decompiled { type, value } recipient as is. Apply the go-live
disclosure to SLA escalation targets too, and give an unattended run a
path: list the recipients in the final summary and do not publish.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013tiDkBvodTad2Je4FVfz4q
… must skip

"Do not publish" named no command, so an agent could read compile as
publishing. Name uip solution upload/publish/deploy run, keep check,
compile and validate running, treat a request that already says to publish
as the user's decision, and count an <UNRESOLVED: ...> target as not
production.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013tiDkBvodTad2Je4FVfz4q
…to an unresolved target

case debug creates real Action Center tasks for the baked-in recipients,
so it joins upload/publish/deploy run. A recipient or escalation target
still <UNRESOLVED: ...> never debugs or publishes; a request that says to
debug or publish decides for a real or test address or group.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013tiDkBvodTad2Je4FVfz4q
upload and publish do not create tasks; only case debug and solution
deploy run do. List non-production recipients in every final summary,
not only before publish, and count a pre-approval that names the debug
or deploy step as the user's answer.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013tiDkBvodTad2Je4FVfz4q

@rockymadden rockymadden left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🟡 Commented: 1 blocker remain.

What

  • Adds case-recipient defaults and disclosures about switching from test recipients to production.

Axes

Axis Blockers Non-blockers Top finding
Correctness & logic 0 0 none
Code quality & simplicity 0 0 none
Types & contracts 1 0 Recipient recipe cannot assign the user · skills/uipath-maestro-case/references/case-runtime.md:53
Tests & validation 0 1 Smoke failure not explained
Security 0 0 none
Architecture & design 0 1 External configuration remains unresolved · skills/uipath-maestro-case/references/case-runtime.md:46
Error handling & resilience 0 0 none
Interface, docs & compatibility 0 0 none

Overall findings

💬 Non-blocker · Tests & validation

The smoke run fails skill-case-edit-add-task-run-once-default: 7/8 tasks pass, below the 95% gate. I could not establish whether this change caused it.

Inspect the failed criterion and fix the regression, or document evidence that it is unrelated.

Validation

  • Checked CA-232, current SDK input/recipient lowering and serialized plan checks, plus flavor inheritance.
  • Tests read, not run. Smoke CI remains red as recorded above.

tl;dr: Repair the recipient recipe; external settings and the smoke failure still need follow-up.

```ts
.input({ approver: types.string })
// ...
.action({ ..., recipient: '=vars.start.output.approver' })

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🛑 Blocker · Types & contracts: use the case recipient contract

This recipe fails before it can route an action: the case entry point does not export types.string. Changing that to 'string' still leaves two problems: case inputs use =vars.approver, and a string recipient is serialized as Type 2, a literal email. =vars.start.output.approver is the Flow path and becomes literal recipient text.

Show the supported input/default syntax and an expression recipient, { type: 3, value: '=vars.approver' }. Replace the script fallback below too: copying the value into another variable does not change recipient Type 2. Current SDK plan checks diagnose this contract.

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fixed in a0e145e, as you suggested. The recipe is a case input plus { type: 3, value: '=vars.approver' }, which compiles and checks clean on SDK 6.120.17. The script fallback is gone. The rule also says that the SDK's RecipientType doc comment, which calls 0/1/3 reserved, is stale.

recipient in the source and republishing the case.

When a request names a person or test mailbox as the recipient, wire it
through a case **input** with a placeholder default instead of a literal.

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💬 Non-blocker · Architecture & design

CA-232 asks for test settings outside the case, such as an asset. This rule still embeds a default and requires a source edit and republish to change it.

Add a supported external configuration route, or explicitly track that ticket requirement as unresolved and describe this change as disclosure guidance.

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Partly addressed in a0e145e. The recipient is now the case input, so going live means changing the input default, or passing the address at start with case process run --inputs. Nobody edits every action any more. An Orchestrator asset as the source is not covered. CA-232 stays open for that part, and the PR body says so.

roberts-cliff and others added 2 commits November 18, 2026 13:47
…xpression

The engine resolves a Type 3 recipient by evaluating its value (CLI
sdd-convert recipient.ts; 68 of 74 corpus expression recipients; TIF-12003
demo). SDK 6.120.17 case check warns STRICT_RECIPIENT_EXPRESSION_TYPE on an
expression under type 0-2 and validate --strict fails it. Teach input +
{ type: 3, value: '=vars.<input>' }, verified by check and compile. Gate any
command that starts an instance, not a fixed verb list.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013tiDkBvodTad2Je4FVfz4q
…tale SDK comment

No evidence shows deploy run or activate starting a manual-trigger case,
so the gate names only case debug and case process run. Say in the rule
that the SDK's RecipientType comment ("0/1/3 reserved") is stale, and
point SKILL.md's literal-recipient example at the go-live section.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013tiDkBvodTad2Je4FVfz4q
@roberts-cliff

Copy link
Copy Markdown
Contributor Author

@dmetzgar Thanks. Your points match tmatup's and Rocky's, and a0e145e fixes them (see the thread replies). There is one change of direction from my earlier rounds. I had dropped the input-wired recipe because I thought runtime support was unconfirmed. That was wrong. The CLI's convert and strict rules already treat { Type: 3, Value: '=vars.x' } as the expression recipient, so the recipe is back in that shape. Per your ask, I did not run case debug. The evidence is the CLI source and an SDK check/compile probe.

Smoke skill-case-edit-add-task-run-once-default (7/8 on the earlier head): not root-caused yet. It tests run-once defaults, not recipients.

@rockymadden rockymadden left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🟢 Approved: recipient contract fixed; 1 non-blocker remains.

What

  • Wires test recipients through a case input and Type 3 expression.
  • Expands assignee disclosures and checks before starting a case.

Axes

Axis Blockers Non-blockers Top finding
Correctness & logic 0 0 none
Code quality & simplicity 0 0 none
Types & contracts 0 0 none
Tests & validation 0 0 none
Security 0 0 none
Architecture & design 0 0 none
Error handling & resilience 0 0 none
Interface, docs & compatibility 0 1 Disclosure happens too late · skills/uipath-maestro-case/references/case-runtime.md:65

Overall findings

  • None.

Previous findings

  • Fixed: supported input/default syntax, =vars.approver, explicit Type 3 recipient, and removal of the script fallback.
  • Addressed: the body explicitly leaves CA-232's external asset requirement open.
  • Cleared: current smoke run passes 8/8, including the previously failing run-once task.

Validation

  • Checked the updated recipe against SDK lowering, CLI recipient rules and engine assignment handling; canonical and Studio Web guidance agree.
  • Source and tests read; no local code, tests or builds run. CI: 51 successful, 8 skipped. No new AI slop found.

tl;dr: The recipe is fixed. Keep the recipient notice before publishing.

Brownfield: keep a decompiled `{ type, value }` recipient as is.

<!-- RULE:case.recipient.no-silent-default -->
- In the final summary of every build or edit, list each assignee the user did

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💬 Non-blocker · Interface, docs & compatibility: disclose before publishing

This replaces the previous instruction to tell the user before publishing. For “build and publish this case,” the final summary can now disclose the test recipient only after publication. CA-232 explicitly asks for the go-live notice before publishing.

State each test assignee and its go-live change before publishing, then repeat that information in the final summary.

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Agreed. CA-232 asks for the notice before publishing, and the rewrite moved it into the final summary only. The fix is one sentence: state each test assignee and its go-live change before upload or publish, then repeat it in the final summary. I haven't pushed it, because this PR is approved and a push would need a new review. It goes in with the next push here, or in a follow-up if this merges first.

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done in 60cac69 and 5451ad2. Before uip solution upload or uip solution publish, the agent now tells the user the list of non-production assignees and what going live changes, then repeats that list in the final summary.

roberts-cliff and others added 2 commits November 18, 2026 14:53
…blish

CA-232 asks for the go-live notice before publishing. Tell the user
before uip solution upload or publish, and repeat it in the final
summary.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013tiDkBvodTad2Je4FVfz4q
@roberts-cliff
roberts-cliff dismissed stale reviews from tmatup and dmetzgar November 18, 2026 21:56

Stale: this review was on f1ba7da. Its blockers were the 'types' import, the Flow path vars.start.output.*, and the claim that case check verifies the recipient. a0e145e fixed them: the recipe is now a case input plus { type: 3, value: '=vars.approver' }, it checks and compiles on SDK 6.120.17, and the strict lint is named. Thread replies give the details. Please re-review 5451ad2.

@tmatup tmatup left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Approve. No blocking issues. My three blocking points are fixed, and the new recipe compiles and passes the skill's own gate. One sentence is wrong about which command catches the bad form (inline, should fix).

I compiled the doc's snippet with the builder SDK and ran uip maestro case validate on the output:

Variant SDK check Emitted recipient validate validate --strict
Doc snippet: .input({ approver: { type: 'string', default: '<TEST_EMAIL>' } }, { from: manual }) + recipient: { type: 3, value: '=vars.approver' } OK {"Type": 3, "Value": "=vars.approver"} Valid Valid
Bare recipient: '=vars.approver' OK {"Type": 2, "Value": "=vars.approver"} Valid Failure: STRICT_RECIPIENT_EXPRESSION_TYPE
  • Fixed:
    • The snippet compiles (no types import). The <TEST_EMAIL> default lands in variables.inputs.
    • The read is =vars.approver, not the Flow-style vars.start.output.* path.
    • Type 3 now has evidence behind it. strict-case-rules.ts on cli main, and its fixture README, record 68 of 74 corpus expression recipients as Type 3 ("Type 3 is Unknown, which the engine routes on the value's shape"). Runtime not verified by me; this is per the CLI's recorded corpus.
  • Caught where it matters: the bare form gets past check and plain validate, but fails validate --strict. SKILL.md:125 makes that the build gate ("complete only when validate --strict passes").
  • Should fix (inline): "case check warns STRICT_RECIPIENT_EXPRESSION_TYPE" is not true. That code exists only in the CLI's strict validate profile (case-validate-service.ts imports strict-case-rules.ts; nothing else does). The SDK's check has no recipient-type rule on main (7f18b8dac), and it reported check: OK for the bare form above.
  • Enabling layer (follow-up, not this PR): the SDK lets the bare form through. .action({ recipient: '<string>' }) always writes Type 2, its RecipientType comment still calls 3 reserved, and its check has no recipient rule. A builder-sdk issue to emit Type 3 for a =-prefixed string (or have check flag it), and to fix the comment, would make this rule unnecessary. I can file it.

Not reviewed: runtime assignment for a Type 3 recipient, and the wording of the disclosure rule beyond reading it.


Review run with Claude Code; each finding was reproduced with a probe at the PR head.

Type `3` evaluates `value` and assigns the result as a user when it is an
email, and as a group name otherwise. An expression needs type `3`: a bare
string `'=vars.approver'` is type `2` and is looked up as a literal email.
`case check` warns `STRICT_RECIPIENT_EXPRESSION_TYPE` on that, and

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should fix: case check does not report this. STRICT_RECIPIENT_EXPRESSION_TYPE is only in the CLI's strict validate profile (strict-case-rules.ts, imported only by case-validate-service.ts). Probe: the bare recipient: '=vars.approver' gives SDK check: OK and plain validate: Valid, and only validate --strict fails with this code. Suggest: "case check and plain validate accept it; case validate --strict fails with STRICT_RECIPIENT_EXPRESSION_TYPE."

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Our probes disagree, and I think it's the SDK version. On @uipath/maestro-builder-sdk@6.120.17 (the current npm release), case-check --source on the bare recipient: '=vars.approver' prints:

⚠ [STRICT_RECIPIENT_EXPRESSION_TYPE] The caseplan breaks a rule `uip maestro case validate` applies, at nodes[…].data.recipient.Type: … with recipient type 2. … an expression recipient is type 3.
check: OK (1 warning(s))

The SDK ports the CLI rule as a lint in dist/case/plan-check/rules/tasks.js (cli: 'strict-case-rules.ts:486', level warning). Older SDKs don't have it; the 6.41.1 on this machine prints a bare check: OK. So on a current SDK the doc is right that check warns, it's a warning not an error, and validate --strict is what fails.

This PR is approved, so I won't push for a non-blocker. Proposed wording for a follow-up: "case check warns STRICT_RECIPIENT_EXPRESSION_TYPE on SDK 6.120 or later (older SDKs print a plain OK), and case validate --strict fails."

@roberts-cliff
roberts-cliff merged commit 5abc161 into main Nov 18, 2026
69 checks passed
@roberts-cliff
roberts-cliff deleted the fix/case-test-recipient-guidance branch November 18, 2026 22:43
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants