controller

name	controller
description	Orchestrates the Quay bug-fix workflow through 9 phases with confidence-based gating. Reads confidence from phase artifacts to advance automatically, post JIRA comments, or escalate to the user.
allowed-tools	["Bash(bash .claude/scripts/session-setup.sh)","Bash(bash .claude/scripts/jira-ops.sh *)","Bash(bash .claude/scripts/tick-state.sh *)","Bash(bash .claude/scripts/format-and-lint.sh *)","Bash(bash .claude/scripts/poll-pr.sh *)","Bash(bash .claude/scripts/validate-pr-title.sh *)","Bash(bash .claude/scripts/validate-commit-msg.sh *)","Bash(bash .claude/scripts/check-ci.sh *)","Bash(git *)","Bash(gh *)","Bash(make *)","Bash(pytest *)","Bash(python *)","Bash(pre-commit *)","Bash(alembic *)","Bash(npm *)","Bash(npx *)","Bash(docker *)","Bash(podman *)","Bash(curl *)","Bash(cat *)","Bash(echo *)","Bash(find *)","Bash(ls *)","Read","Write","Edit","Glob","Grep","Agent","AskUserQuestion","TodoWrite","CronCreate","CronDelete","CronList"]

Quay Bugfix Controller

You manage a 9-phase bug-fix workflow with confidence-based gating. After each phase, read the confidence assessment from the phase artifact and use it to decide whether to advance, comment, or escalate.

Session Bootstrap

On first run, ensure Lola plugins are installed:

bash .claude/scripts/session-setup.sh

Phases

1. Assess — the assess skill Read the bug report, summarize understanding, identify gaps, propose a plan.

2. Reproduce — the reproduce skill Confirm the bug exists by reproducing it in a controlled environment.

3. Diagnose — the diagnose skill Trace the root cause through code analysis, git history, and hypothesis testing.

4. Fix — the /dev:code skill (from dev plugin) Read the root cause analysis, create a feature branch, then implement the minimal fix using /dev:code. Write implementation notes afterward.

5. Test — the test skill Write regression tests, run the full suite, and verify the fix holds.

6. Review — the review skill Critically evaluate the fix and tests — look for gaps, regressions, and missed edge cases.

7. Document — the document skill Create release notes, changelog entries, JIRA updates, and PR description.

8. PR — the /dev:pr skill (from dev plugin), then /dev:poll Create a pull request using /dev:pr, then start CI polling with /dev:poll <PR#>.

9. Summary — the summary skill Scan all artifacts and present a synthesized summary.

Confidence-Based Gating

Confidence Assessment Format

Each phase skill writes a ## Confidence Assessment section at the bottom of its artifact:

## Confidence Assessment
- **Level**: high | medium | low
- **Score**: <0-100 integer>
- **Score rationale**: <1-2 sentences>
- **Open questions**: <bullet list, or "None">

Confidence Flow

After each phase completes, read the confidence level from the artifact:

Confidence	Threshold	Action
High	>=90%	Advance to next phase silently
Medium	70-89%	Post JIRA comment with findings and open questions, then advance
Low	<70%	Post JIRA comment, then stop and escalate via AskUserQuestion

Posting JIRA Comments

When confidence is medium or low, post a structured comment:

bash .claude/scripts/jira-ops.sh comment <TICKET_KEY> "<comment_text>"

Format the comment as:

[Phase: <phase_name>] Automated Analysis Update

Confidence: <Level> (Score: <score>/100)

Findings:
- <key finding 1>
- <key finding 2>

Open Questions:
- <question 1>
- <question 2>

Next: <what the agent will do next, or "Stopping for human input">

JIRA Ticket Context

The controller needs a JIRA ticket key to post comments. The ticket key comes from:

1. The assess phase (extracted from the bug report or user input)
2. Direct user input

If no ticket key is available, skip JIRA comments entirely and use AskUserQuestion for medium-confidence escalations too.

How to Execute a Phase

1. Announce the phase to the user before doing anything else.
2. Run the skill for the current phase.
3. Read the confidence assessment from the phase artifact.
4. Act on the confidence level per the table above.
5. If advancing, continue to the next phase.

Recommending Next Steps

After each phase, log the natural next step. These are informational — the agent advances automatically based on confidence, not user choice.

• After assess: Next is reproduce (or skip to diagnose if root cause is already evident from the report).
• After reproduce: Next is diagnose (or skip to fix if reproduction confirmed the cause).
• After diagnose: Next is fix (or re-assess if diagnosis revealed a different bug).
• After fix: Next is test. Always test before PR.
• After test: Next is review.
• After review:
  • Verdict "solid" -> document (then auto-advance through PR and summary)
  • Verdict "tests incomplete" -> test (add missing coverage)
  • Verdict "inadequate" -> fix (address review concerns, max 2 cycles)
• After document: Next is pr.
• After pr: Next is summary.

Escalation Rules (Override Confidence)

These conditions always trigger escalation via AskUserQuestion, regardless of the phase confidence level:

• Security vulnerability discovered
• Multiple valid solutions with unclear trade-offs
• Architectural decisions that affect other teams
• Existing PR or fix found for the same issue (assess phase)
• Review verdict "inadequate" after 2 revision cycles

Starting the Workflow

When the user provides a bug report, issue URL, or JIRA ticket:

1. Execute the assess phase
2. Read the confidence assessment
3. Continue per the confidence flow

If the user invokes a specific skill directly, execute that phase — don't force them through earlier phases.

Rules

• One mode. The controller always uses confidence-based gating.
• Recommendations come from this file, not from skills. Skills report findings and confidence; this controller decides what to do next.
• Max 2 revision cycles. If review says "inadequate" twice, stop and escalate regardless of confidence.