Skip to content

feat(uipath-platform): teach agents to write and query trace decisions - #4025

Open
dushyant-uipath wants to merge 1 commit into
mainfrom
feat/decisions-write-skill
Open

dushyant-uipath wants to merge 1 commit into
mainfrom
feat/decisions-write-skill

Conversation

@dushyant-uipath

@dushyant-uipath dushyant-uipath commented Oct 16, 2026 •

Copy link
Copy Markdown
Collaborator

Problem

An agent asked to record a reviewer's decision on a trace has no skill guidance. The Decisions API needs a trace ID, a folder key, and a request body where every decided field carries a proposer. Without a documented command the agent builds a raw HTTP call and guesses the body.

The commands this teaches are added in UiPath/cli#5280 (uip traces decisions create|get|list), which calls the Decisions service added to the TypeScript SDK in UiPath/uipath-typescript#827 (merged, released in SDK 1.8.4). The CLI change is not merged yet, so uip traces decisions is not in a released CLI until it lands.

Change

uipath-platform gets a decisions reference next to the existing trace feedback reference. No new skill folder, so assets/skill-status.json, skills.sh.json, CODEOWNERS and the README table stay as they are.

File Change
skills/uipath-platform/references/traces/decisions.md New. Critical rules, create flag table, input file shape, get, list, error table, anti-patterns
skills/uipath-platform/SKILL.md when_to_use gains the decision phrases, the Traces bullet, the routing table row and the reference list entry link the new file
skills/uipath-platform/references/traces/traces.md Related link
skills/uipath-platform/references/uip-commands.md Three command rows under Traces
tests/tasks/uipath-platform/traces/traces_decisions_smoke.yaml Smoke task graded on the argv of traces decisions create and traces decisions list against a recorded uip mock
flowchart LR
    A[User asks to record a decision] --> B[uipath-platform SKILL.md]
    B --> C[references/traces/decisions.md]
    C --> D[uip traces decisions create]
    C --> E[uip traces decisions list / get]
    D --> F[Decisions service - uipath-typescript#827]
    E --> F
    F --> G[llmops Decisions API]
Loading

Tradeoffs

Option How it works Problem Verdict
Reference inside uipath-platform Sits beside traces/feedback.md, same CLI tool and same backend when_to_use grows by one phrase Chosen
New uipath-decisions skill Own SKILL.md, status entry, grouping, CODEOWNERS, README row Three registry edits and a second skill competing for "trace" requests Rejected
Add to traces/feedback.md One file for both Feedback and decisions have different required fields and a reader of one loads the other Rejected
Docs only, no smoke task Cheaper Nothing checks that an agent reaches the right verb and flags Rejected

description is unchanged because it sits 2 characters under the 1024 cap; the new trigger phrases go in when_to_use. The CLI verb catalog does not list traces decisions until the CLI change is released and the catalog refresh runs.

[ACTN-0000]

@github-actions

github-actions Bot commented Oct 16, 2026 •

Copy link
Copy Markdown
Contributor

Claude finished @dushyant-uipath's task in 2m 56s —— View job


PR Review: feat(uipath-platform): teach agents to write and query trace decisions

  • Gather review criteria and project conventions
  • Read full diff and all changed files
  • Validate structure, links, and content quality
  • Check test task compliance
  • Post review findings

Summary

Adds a decisions.md reference alongside the existing traces/feedback.md inside uipath-platform, teaching agents the uip traces decisions create|get|list commands. SKILL.md gains trigger phrases, a routing-table row, and a reference-list entry. A record_cli-based smoke test verifies argv for create and list. Well-scoped — no new skill folder, so registry files stay untouched.

Change-by-Change Review

1. skills/uipath-platform/references/traces/decisions.md (new)

Severity: Low

Clean reference that mirrors the structure of feedback.md. Critical Rules numbered, anti-patterns present, error table present, CLI examples use --output json, placeholders consistently use <UPPER_SNAKE_CASE>. Two minor observations:

  • list required flags unclear (line 117-134): The create flag table explicitly marks --trace-id and --folder-key as "Yes" required, but the list flag table has no "Required" column — an agent won't know whether --trace-id or --folder-key is mandatory or optional for list. The test asserts --trace-id on list, and the example includes --folder-key, but the table is silent. Compare with feedback.md line 67 which explicitly says --folder-key: Optional. Fix this →

  • decisionId required in JSON but optional via flags (line 102 vs. critical rule feat(CodedAgents): single skill #3): The input-file-shape section states decisionId is required per decision, while critical rule feat(CodedAgents): single skill #3 says "Omit --id to get a generated one." If this is by design (server generates only for flag form, not file form), a one-line note in the input-file section would prevent agents from omitting it in JSON and hitting a 400. Fix this →

2. skills/uipath-platform/SKILL.md (+4/−2)

Severity: OK

  • when_to_use gains 'write / record / query a decision on a trace' — concise and hits the right trigger phrases.
  • Traces bullet updated accurately: create|get|list.
  • Routing table row added in the right position (after Feedback, before IS).
  • Reference list entry added with correct relative link.
  • description unchanged (still under 1024).

All relative links verified — references/traces/decisions.md exists on disk.

3. skills/uipath-platform/references/traces/traces.md (+1)

Severity: OK

One Related link added, matching the pattern of the existing Feedback link. Link target verified.

4. skills/uipath-platform/references/uip-commands.md (+4/−1)

Severity: OK

Three command rows added under the Traces table. Descriptions are concise and accurate. Section header text updated to mention "decisions." Link to traces/decisions.md verified.

5. tests/tasks/uipath-platform/traces/traces_decisions_smoke.yaml (new)

Severity: OK

Follows the established traces_feedback_smoke.yaml pattern closely:

  • task_id follows skill-platform-traces-decisions-smoke convention.
  • Tags include uipath-platform first, smoke, and mode:build — valid vocabulary.
  • Uses record_cli mock with a 404 stub — same approach as sibling traces tests.
  • Three cli_called criteria with sensible weights (3.0 primary, 1.5 supporting).
  • outcome: {present: true, aliases: [input]} correctly accepts both flag form (--outcome) and file form (--input).
  • No e2e task — acceptable since the CLI verbs (uip traces decisions) aren't released yet (UiPath/cli#5280). Smoke coverage is the right starting point.

What's Missing

  • list flag table lacks a Required column — see finding 1 above. Low severity but could cause agents to omit required flags or pass unnecessary ones.
  • No e2e test — expected, since the CLI commands haven't shipped yet. Should be added once uip traces decisions is in the catalog.

Area Ratings

Area Status Notes
Frontmatter OK description unchanged (under 1024), when_to_use updated correctly
E2E Tests OK Smoke test present; e2e deferred until CLI ships (acceptable)
Skill Body OK Routing table, trigger phrases, reference list all updated
References & Assets OK (minor) New reference is well-structured; list required-flags clarity is the only gap
Repo Hygiene OK No new skill folder — registries/CODEOWNERS untouched. No secrets or personal paths. All links resolve

Issues for Manual Review

  • CLI flag accuracy: The create flag table lists --field, --proposed-by, --proposed-by-kind, --rejected-outcome, --evidence, --timestamp, --process-key, etc. These come from UiPath/cli#5280 which hasn't shipped — verify the actual flag names match once the CLI is released.
  • decisionId generation behavior: Confirm whether the server auto-generates decisionId only for flag-form input or also for JSON file input. The doc currently says "required" in the file shape but "optional" via --id.

Conclusion

Clean, well-scoped addition that follows the existing traces reference pattern. The two Low-severity findings (list-flag clarity and the decisionId required-vs-optional note) are polish items — neither blocks an agent from using the commands correctly. Approve.


|

@github-actions

github-actions Bot commented Oct 16, 2026 •

Copy link
Copy Markdown
Contributor

Claude finished @dushyant-uipath's task in 3m 0s —— View job


Coder-eval task lint (advisory)

  • Read lint rubric
  • Identify changed task YAMLs
  • Check PR body for evidence of passing run
  • Apply rubric to each changed task
  • Post final comment

Coder-eval task lint (advisory)

1 task YAML changed; verdict counts: 0 Critical, 1 High, 0 Medium, 0 Low, 0 OK.

Rubric: .claude/commands/lint-task.md. This check is advisory and never blocks merge.

Evidence of passing run

❌ High — PR body does not claim the changed task has been run and passed. Please edit the PR description to add a line like: Ran skill-platform-traces-decisions-smoke locally and it passed.

Per-task lint

tests/tasks/uipath-platform/traces/traces_decisions_smoke.yaml — verdict: High

Issues:

  • [High] CLI verb reachability: verbs traces decisions create (lines 35, 47) and traces decisions list (line 54) are not present in assets/uip-catalog-snapshot.json. The cli_called criteria reference verbs that don't exist in the catalog yet. PR body acknowledges this ("The CLI verb catalog does not list traces decisions until the CLI change is released and the catalog refresh runs") — once the catalog refreshes via .github/workflows/refresh-uip-catalog.yml this finding will clear itself.

Suggested fixes:

  • After uip ships traces decisions create|get|list (UiPath/cli#5280) and the nightly catalog refresh runs, verify the verbs appear in assets/uip-catalog-snapshot.json. If the task needs to merge before the refresh, run python3 scripts/build-uip-catalog.py locally to update the catalog snapshot.

Within-PR duplicates

  • No duplicate clusters detected (only 1 new task YAML).

Conclusion

⚠ 1 task has issues, max severity High. Advisory only — not blocking merge. The CLI verb gap is expected pre-release and self-resolves once the CLI ships. The missing passing-run claim is the actionable item — please add it to the PR description.


@dushyant-uipath
dushyant-uipath force-pushed the feat/decisions-write-skill branch from 3cebf96 to e62e710 Compare October 19, 2026 09:57
@dushyant-uipath

Copy link
Copy Markdown
Collaborator Author

Addressed both review comments in skills/uipath-platform/references/traces/decisions.md, amended into the existing commit (e62e710):

  1. Added a Required column to the list flags table (all flags there are optional filters, confirmed against the API's OpenAPI schema).
  2. Clarified that decisionId is required in every decision object of a --input JSON file, unlike the flag form where omitting --id generates one.

@sonarqubecloud

Copy link
Copy Markdown

@dushyant-uipath
dushyant-uipath force-pushed the feat/decisions-write-skill branch from e62e710 to 24fda69 Compare November 13, 2026 08:28
Adds references/traces/decisions.md for uip traces decisions create, get and list, links it from SKILL.md, traces.md and uip-commands.md, and adds a smoke task.

Addressed review feedback: added a Required column to the list flags table; clarified that decisionId is required in --input files, unlike the flag form which generates one when --id is omitted.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

This branch has not been deployed

No deployments
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.

1 participant