aurora

When something in AURORA broke and you want to tell us.

Validate `policy.yaml` and run policy-shape dry-runs. Hard fails on schema errors or unresolved env vars; warns on risky configurations. Dry-run shows what Conductor would do over the next N hours given the current backlog and Operate-fleet state, without dispatching anything.

Search AURORA's three-tier memory. Project tier (per-bot artifacts), org tier (cumulative patterns and quirks), skill tier (one-line learnings and the fingerprint index). Returns ranked excerpts with paths so you can drill in. The same retrieval the agents use internally — visible to the human.

Boot the swarm.

Live dashboard for the swarm.

Discovery-fleet PDD author and ROI scorer. Reads a pending candidate from the backlog, drafts a structured Process Definition Document, and computes an ROI score = frequency × pain × feasibility, weighted per `policy.yaml`. Decides whether ambiguity is high enough to require Interviewer input. Use this agent after Curator promotes a candidate to `pending-analyst`.

Build-fleet pattern selector. Reads a PDD with low ambiguity (status `ready-for-architect`) and writes an Architecture Decision Record specifying which UiPath pattern to use — Sequence, REFramework Performer, REFramework Dispatcher, Coded Workflow (C#), Coded Agent (LangGraph / OpenAI Agents / LlamaIndex), Maestro, Action Center, API Workflow, or Document Understanding. Use this agent after Analyst (or Interviewer) sets a candidate to `ready-for-architect`. Architect is the only agent that picks patterns; Conductor never invokes a Forger directly.

Operate-fleet governance and drift checker. Compares repo XAML/coded-workflow hashes against deployed package contents, reconciles license utilization, identifies idle processes (deprecation candidates), and runs the pre-promote governance pass before any prod deploy. Use this agent before promoting to prod, when Strategist proposes a deprecation, after a Surgeon-led redeploy, or on a configurable cadence.

Build-fleet UI explorer. Builds the UiPath Object Repository for any process whose ADR includes UI automation. Drives Playwright MCP for web targets and `inspect-ui-tree.ps1` for Windows desktop targets. Captures strict (single-find) selectors covering all relevant technologies — wnd, html, webctrl, aa, uia, java, sap. Outputs `.objects/` tree and `references.json`. Use this agent when ADR forgers list includes any UI-driven workflow. Runs in parallel with the Forger sub-fleet.

Operate-fleet bridge between the always-running swarm and asynchronous humans. Creates UiPath Action Center Form Tasks (or App Tasks) via the uipath-python SDK, polls for completion, and routes the human's response back into the calling agent's state. Owns the implementation of every HITL gate from `policy.yaml::gates`. Use this agent whenever any peer needs human input — Interviewer's questions, Conductor's prod-publish gate, Strategist's deprecation, Surgeon's large-fix gate, the compost-step skill PR.

Mint and refresh UiPath Automation Cloud OAuth tokens via the External Application client-credentials grant. Reads UIPATH_CLIENT_ID and UIPATH_CLIENT_SECRET from .env, requests the scope list from policy.yaml::uipath_scopes, writes the live access token to UIPATH_ACCESS_TOKEN so the uipath CLI can use it without re-auth, and refreshes proactively before the 1-hour expiry. Use this skill at the start of every Conductor run, before any uipath CLI or uipath-python SDK call, and whenever Sentry emits a `kind: auth_failed` event.

Nightly self-improvement loop. Reads the day's `learnings/<date>.jsonl`, clusters them by skill/agent, identifies recurring patterns (≥ 3 occurrences across ≥ 2 projects), and opens a GitHub PR against `skills/` (or `agents/`) with a proposed update — e.g., a new fingerprint refinement, an updated SKILL.md guidance, a new helper script. The PR is HITL-gated via `aurora-promote` (kind: skill_compost_pr) and never auto-merged. The mechanism that makes the swarm get smarter with use.

Retire an unused or superseded UiPath process safely. Used by Strategist + Auditor + Concierge in sequence: Strategist proposes, Auditor checks dependencies, Concierge gates via Action Center, then this skill executes the deprecation steps. Stops scheduled triggers, archives the package, transfers any in-flight queue items to a successor (if any), updates org memory's deprecation log, and writes a runbook so the retired process can be restored within 30 days if needed.

Extract structured friction signals from raw text sources (Slack messages, Jira issues, email digests, calendar invite notes, screen-recording transcripts). Used by `scout` to convert noisy human chatter into machine-actionable candidate signals. Returns one JSON object per signal with frequency hints, duration hints, pain indicators, and named systems. Reject one-off, internal-swarm, and metadata-only chatter.

Cluster failure events by structural fingerprint and append to memory. Used by Diagnostician and Surgeon. A fingerprint is `<top-level-kind>/<refinement>` plus locality (workflow file or agent name). Clustering uses kNN over a structured-feature embedding stored in SQLite. Output is a cluster ID and confidence score; downstream agents use these to decide auto-fix vs HITL.