workflow-demos

Drive a task end-to-end from description to opened PR (Plan → Implement → Review → Fix → PR). Recreation of the built-in autopilot workflow via subagents. MODIFIES code.

Reproduce, fix, and verify a single reported bug via subagents. Recreation of the built-in bugfix workflow (native Workflow tool is gated off fleet-wide).

Adversarial bug hunt over the current branch (or a given path/diff-range). Self-respawning finder fleet + 5-vote refutation jury. Recreation of the built-in bughunt workflow via subagents.

Fast fixed-fleet bug hunt of the current branch (or a given path/diff-range). Recreation of the built-in bughunt-lite workflow via subagents — lighter/faster than bughunt.

Build a working dashboard from a plain-language request via a 5-phase subagent pipeline (Discover → Design → Implement → Verify → PR). Recreation of the built-in dashboard workflow. CREATES + commits files and opens a PR.

End-to-end "task → merged PR" build loop driven by subagents. Takes a task description and runs Plan (draft + 5 parallel critics + harden) → Implement → Review (3 rapid + 2 deep bughunters + adversarial verify, parallel with a completeness check) → Fix → PR. Use for "autopilot this", "build this feature end to end", "implement and open a PR for X", "drive this task to a PR", or the /autopilot-demo slash command. Faithful recreation of the Claude Code built-in `autopilot` workflow using subagents — works WITHOUT the gated Workflow tool (`tengu_workflows_enabled` is OFF fleet-wide). NOTE: this MODIFIES code (it is a build workflow, not review-only).

End-to-end single-bug fix loop driven from a bug report (Reproduce → Root-cause → Fix → Regress → PR). Dispatches a strictly sequential chain of 5 subagents — write the smallest failing repro, trace to the minimal root cause, apply the minimal fix and re-run the repro, harden the repro into a permanent regression test and run the suite, then lint/typecheck/branch/commit/push and open a PR. Hard-gates between phases (no repro → stop; fix not done → stop). Use for "fix this bug", "reproduce and fix", "here's a bug report — fix it", or the /bugfix-demo slash command. MODIFIES code. Faithful recreation of the Claude Code built-in `bugfix` workflow using subagents — works WITHOUT the gated Workflow tool (tengu_workflows_enabled is OFF fleet-wide).

Fast, fixed-fleet bug hunt over a branch diff (or a specified path/diff-range). Dispatches a FIXED fleet of 5 finders (3 rapid surface scanners + 2 deep analysts) in parallel, runs 5-voter adversarial verification per candidate (2 refutations kill a bug; early-kill after the first 2 votes), naively dedups by file:line, then synthesizes a semantically-deduped, severity-grouped report. Lighter and faster than the full `bughunt`: no self-respawning slots, no dry-streak loop, no skip-key feedback — just one fixed wave of finders. Use for "quick bug hunt", "bughunt-lite", "fast bug scan of my branch", "hunt bugs before PR (light)", or the /bughunt-lite-demo slash command. Faithful recreation of the Claude Code built-in `bughunt-lite` workflow using subagents — works WITHOUT the gated Workflow tool (tengu_workflows_enabled is OFF fleet-wide).

Adversarial bug hunt over a branch diff (or a specified path/diff-range). Runs a self-respawning finder fleet (rapid surface scanners + deep analysts) that dispatches bug candidates the moment they're found, verifies each candidate with a 5-vote pigeonhole adversarial jury (≥2 refutations kill it), and synthesizes a semantically-deduped, severity-ranked report. Terminates on a deep-finder dry streak. Use for "hunt for bugs", "find bugs in my branch", "bug hunt", "adversarial bug review", or the /bughunt-demo slash command. Faithful recreation of the Claude Code built-in `bughunt` workflow using subagents — works WITHOUT the gated Workflow tool (tengu_workflows_enabled is OFF fleet-wide).

Build a working dashboard from a plain-language request by running a 5-phase pipeline — Discover the dashboard stack + data sources, Design the panels, Implement the dashboard file(s), Verify queries render, then open a PR. Each phase is one subagent returning strict JSON. Use for "build a dashboard for X", "make a Grafana/Hex/Streamlit/React dashboard", "add a metrics dashboard", "dashboard the <data>", or the /dashboard-demo slash command. Faithful recreation of the Claude Code built-in `dashboard` workflow using subagents — works WITHOUT the gated Workflow tool (tengu_workflows_enabled is OFF fleet-wide). This skill CREATES and commits files.