This repository contains self-contained AI agent skills for UiPath automation development. Skills are installed as a Claude Code plugin and teach AI agents how to build, run, test, and deploy UiPath automations.
- Skills are self-contained. Each skill under
skills/must function on its own: it MUST NOT import, inline, or read another skill's files, and MUST still deliver its core value if sibling skills are absent. A skill MAY delegate a task to a sibling skill in this same plugin at runtime (e.g., spawn a subagent that hands an artifact edit to the artifact's owning domain skill) when that work is the sibling's domain — provided the delegation degrades gracefully (the skill still presents an actionable result when the sibling is unavailable). - SKILL.md is the canonical default contract. Every skill folder must have a complete
SKILL.mdwith valid YAML frontmatter. Local/default consumers can understand it without a flavor manifest or runtime resolver. Shared guidance stays there; only genuinely different passages may be enclosed by named flavor-block comments. - Custom flavors inherit the complete catalog and contain only sparse exceptions. Every flavor ships every canonical skill. Files under
skill-flavors/<flavor>/<skill>/mirror canonical skill paths and contain only complete replacement blocks. If no override exists, the canonical file is intentionally reused unchanged. Do not duplicate a whole skill merely to change a few paragraphs. - Build files before packages. Build complete
defaultand custom-flavor skill trees, validate those final files, and only then stage packages. Consumers must copy a finished tree; they must not interpret flavor markers or compose Markdown at runtime. - Session hooks are Node scripts or direct
uipingestion — never shell scripts. Telemetry ships NO script:hooks/hooks.jsonpipes the raw hook payload touip track --hook, and the field derivation lives in the CLI (UiPath/clitrack-hook.ts, guarded by its vitest suite). The remaining session hooks are each onehooks/<name>.mjsrun bynode. Bothnodeanduipare native-executable entry points, so no shell dialect and no PowerShell execution policy applies, and both are present wherever the skills were installed through the Node-baseduipCLI. Eachhooks.jsonentry is a bash/PowerShell polyglot command whose both branches guard for the missing binary (silent exit 0) — telemetry's PowerShell branch resolvesuip.cmdexplicitly, never the bareuip(PowerShell would prefer npm'suip.ps1shim, which execution policy can block). Do not add ashellfield to these entries; the polyglot depends on Claude Code's default shell selection. Do NOT reintroduce.sh/.ps1hook implementations: the retired PowerShell twins required an Authenticode signing gate at publish and still failed under Group-Policy execution policy.
See CONTRIBUTING.md for the full guide. Key rules:
-
Skill folder naming:
uipath-<kebab-case>underskills/ -
SKILL.md frontmatter is required: must include
name(matching folder name) anddescription(with TRIGGER/DO NOT TRIGGER conditions) -
References use kebab-case filenames with
-guide.mdand-template.mdsuffixes -
Update CODEOWNERS when adding or modifying skill ownership
-
Update the two skill registries in the SAME PR that adds, renames, or removes a skill folder. Adding or renaming
skills/<name>/requires the matching edit toassets/skill-status.json(lifecycle status) AND toskills.sh.json(display grouping); removing a skill requires deleting its entry from both. Neither file is discovered from disk, so nothing else in the build notices when they drift. Verify both before opening the PR:python3 scripts/check-skill-status.py --write-readme # regenerates the README table python3 scripts/check-skills-sh.py # add/rename/remove: must print OK
Full change→edit table, the
--fixlimits, and how pre-existing drift is reported:.claude/rules/skill-structure.md§ skills.sh Grouping. -
No structural cross-skill dependencies — a skill must work in isolation (never import or read another skill's files); runtime delegation to a same-plugin sibling skill is allowed when it degrades gracefully
-
No secrets or personal paths in committed files
-
CLI commands must use
--output jsonwhen output is parsed programmatically -
Review new skills for every custom flavor. They are included automatically; add the smallest sparse override wherever canonical guidance is not safe for a target environment
-
Never rename a job that produces a required status check, and never narrow its trigger. A required context is a literal string GitHub waits for. Rename the job and the context stops reporting; add
paths:,paths-ignore:,branches:,branches-ignore:or atypes:list withoutsynchronizeand the workflow won't run on excluded PRs. Either way the check never arrives and every open PR blocks, not just yours. Short-circuit inside the job instead — a skipped job counts as a pass. The set lives indocs/REQUIRED-CHECKS.md§ Current target set and is not discovered from disk; edit it in the same PR as the workflow:python3 -m pytest tests/scripts/test_required_checks_contract.py -q # must pass ./scripts/apply-required-checks.sh --dry-run # preview the ruleset ./scripts/apply-required-checks.sh --check # doc vs live ruleset (needs repo admin)
| File | Convention |
|---|---|
SKILL.md |
Required. Uppercase. YAML frontmatter + markdown body. |
references/*.md |
Kebab-case. Guides end with -guide.md. |
skill-flavors/<flavor>/<skill>/**/*.md |
Optional sparse overrides at paths relative to skills/; contain only named replacement blocks. |
<!--skill-flavor:<name>:start|end--> |
Compact, whitespace-free column-one boundary around the smallest canonical passage that differs by flavor. Names are lowercase kebab-case and unique within a file; indent the enclosed Markdown, never the boundary. |
assets/templates/* |
Templates end with -template.md or -template.<ext>. |
skills.sh.json |
Display grouping for the repo's skills.sh page. Every skills/<name>/ appears in exactly one grouping. Edit it in the same PR that adds, renames, or removes a skill folder — validate with python3 scripts/check-skills-sh.py. |
assets/skill-status.json |
Lifecycle status (stable / preview / in-development) for every skill — the single source of truth. Edit in the same PR as the skill folder; regenerate the README table with python3 scripts/check-skill-status.py --write-readme. |
docs/REQUIRED-CHECKS.md |
The required-status-check set, plus the rules that keep a context reportable. Machine-read: scripts/parse-required-checks.py is its only parser, shared by scripts/apply-required-checks.sh and the contract guard. Edit § Current target set in the same PR that renames a required job or changes its trigger. |
.skill-lint.yaml |
Severities and options of the skill-lint PR check (.github/workflows/skill-lint.yml). Read from the base branch, so an edit applies once merged. See CONTRIBUTING.md § skill-lint PR check. |
hooks/*.mjs |
Non-telemetry session hooks are single Node.js (>= 20) scripts, spawned by the polyglot commands in hooks/hooks.json. Telemetry has no script — its hooks.json entries pipe the raw payload to uip track --hook. Everything is never-fail by contract (exit 0); the polyglot guards for a missing node/uip and silently no-ops. |
- Before changing flavor markers, overrides, discovery, package construction, publishing, or flavor CI, read
.claude/skills/manage-skill-flavors/SKILL.mdcompletely and follow its reference-routing instructions - Read the existing SKILL.md before making changes
- Read every matching flavor override before changing a canonical marked passage; prefer an empty additive extension block over replacing a shared table, list, or navigation section
- Preserve the Critical Rules section — these prevent expensive agent mistakes
- Validate YAML frontmatter — broken frontmatter breaks skill discovery
- Ensure
descriptionfield has both TRIGGER and DO NOT TRIGGER conditions - When canonical flavor blocks or custom overrides change, run
npm run skills:validate,npm run skills:build, andnpm run skills:pack; inspect the actual generated tarballs before publishing - Preserve normal root
npm packandnpm publishas backward-compatible default-only commands: they must compose marker-free skills transactionally and restore canonical sources. Keepnpm run skills:packas the all-flavor build/verification command used by CI and isolated flavor publishers, not as the default release path
Tests live in tests/tasks/<skill-name>/ as coder_eval task YAMLs. Before authoring or editing a task, read tests/README.md for the full framework: tag taxonomy, experiment configs, success-criteria types, weight guidance, and the /generate-task and /test-coverage slash commands. Repo-specific authoring constraints (workflow, required tags, sandbox rules, anti-patterns) are in .claude/rules/test-writing.md.
- Codereval dashboard results are stored in Azure Blob Storage account
coderevaltests, containerruns. - Run IDs are timestamped like
YYYY-MM-DD_HH-MM-SS; list today's runs with prefixYYYY-MM-DD. - The run summary is
<run-id>/run.json; per-task results live under<run-id>/<variant>/<task-id>/<replicate>/task.json.