Claude-skill-registry agent-ops-focus-scan

Analyze issues to identify the next work item and update focus.md. Enforces issue-first workflow and confidence-based batch limits.

install
source · Clone the upstream repo
git clone https://github.com/majiayu000/claude-skill-registry
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/majiayu000/claude-skill-registry "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/data/agent-ops-focus-scan" ~/.claude/skills/majiayu000-claude-skill-registry-agent-ops-focus-scan && rm -rf "$T"
manifest: skills/data/agent-ops-focus-scan/SKILL.md
source content

Focus Scan

Purpose

Align session focus with the highest priority pending issue. Enforces that all work is tracked as issues and respects confidence-based batch limits.

Context Optimization

Read

index.md
first for quick issue overview. Only load full priority files when selecting specific issues to work on.

1. Read .agent/issues/index.md (compact summary)
2. Identify target priority level from index
3. Load only that priority file for full details

CLI Commands

Works with or without

aoc
CLI installed. Focus scanning can be done via direct file reading.

File-Based Priority Scan (Default)

1. Read .agent/issues/index.md (compact summary)
2. Identify target priority level from index
3. Load only that priority file for full details
4. Filter for status: todo or open
5. Select first actionable issue (respecting depends_on)

Focus Update (File-Based)

  1. Read current
    .agent/focus.md
  2. Update "Doing now" section with selected issue
  3. Update "Next" section with immediate step

CLI Integration (when aoc is available)

When

aoc
CLI is detected in
.agent/tools.json
, these commands provide convenience shortcuts:

CommandPurpose
aoc issues list --status todo,open --priority critical
Find critical actionable issues
aoc issues list --status todo,open --priority high
Find high priority issues
aoc issues list --status blocked
Find blocked issues
aoc issues show <ID>
Check issue details
aoc issues summary
Quick status overview

Confidence-Based Batch Limits

ConfidenceMax Issues per IterationRationale
LOW1 (hard limit)High uncertainty requires focused human oversight
NORMALUp to 3Standard workflow with reasonable batching
HIGHUp to 5Well-understood work can be batched

Enforcement:

  • Read confidence from constitution or current task
  • If LOW confidence, select EXACTLY 1 issue — no batching
  • Present batch to user for confirmation if confidence is not HIGH

Issue-First Enforcement

Before starting any work:

  1. Check for existing issue: Is there an issue for the requested work?

    • Yes → proceed to priority scan
    • No → prompt to create issue first
  2. If user requests work without an issue:

    ⚠️ No issue found for this work.
    
    All work should be tracked for auditability.
    
    Create an issue first?
    [Y]es, create and start / [N]o, work without tracking
    
    Suggested: {TYPE}-{next}@{hash} — "{inferred title}"
    
  3. Even simple chores need issues:

    • "Fix typo" →
      CHORE
      issue
    • "Update dependency" →
      CHORE
      issue
    • "Add comment" →
      DOCS
      issue
    • "Quick refactor" →
      REFAC
      issue (needs permission anyway)

Procedure

  1. Check confidence level (from constitution or task):

    • Determine batch size limit
    • Note if LOW confidence (special handling)
  2. Scan priority files in order (stop when actionable issue found):

    • .agent/issues/critical.md
      (scan first — highest priority)
    • .agent/issues/high.md
    • .agent/issues/medium.md
    • .agent/issues/low.md
      (only if nothing higher)
    • SKIP
      .agent/issues/backlog.md
      — backlog items need triage first
  3. Identify the highest priority issue where:

    • Status is
      todo
      or
      in_progress
    • All
      depends_on
      dependencies are
      done
      (or list is empty)
    • Issue is not marked as
      blocked
      ,
      cancelled
      , or
      dropped
  4. Apply batch limit based on confidence:

    For LOW confidence:

    🎯 LOW CONFIDENCE MODE — Single Issue Focus
    
    Selected: {ISSUE-ID} — {title}
    
    Batch size: 1 (hard limit for low confidence)
    
    This issue will receive:
    - Mandatory interview before planning
    - Extensive implementation details
    - Hard stop after implementation for human review
    - ≥90% test coverage requirement
    
    Proceed with this issue? [Y]es / [N]o, select different
    

    For NORMAL confidence:

    📋 Selected {N} issues for this iteration (max 3):
    
    1. {ISSUE-ID-1} — {title}
    2. {ISSUE-ID-2} — {title}
    
    Proceed? [Y]es / [M]odify selection / [S]ingle issue only
    

    For HIGH confidence:

    • Proceed with selection (up to 5), report only
  5. Action based on findings:

    If an actionable issue is found:

    • Read
      .agent/focus.md
    • Update
      Doing now
      to reflect this issue (include issue ID)
    • Update
      Next
      with the immediate practical step
    • Example:
      ## Doing now
      - Working on FEAT-0012@a3f2c1: Add user authentication
      
      ## Next
      - Review existing auth patterns in codebase
      - Create implementation plan
      

    If all issues are blocked:

    • Identify the specific blockers
    • Update focus to indicate blockers need resolution
    • Suggest resolving dependencies or creating new issues

    If no prioritized issues exist but backlog has items:

    • Prompt user:
      📋 No prioritized issues, but {N} items in backlog.
      
      1. Triage backlog (/agent-task triage)
      2. Create new issue (/agent-task)
      3. Run discovery (/agent-baseline, /agent-review)
      

    If no issues exist at all:

    • Update focus to indicate planning/discovery needed
    • Prompt user:
      📋 No issues anywhere. What would you like to do?
      
      1. Create new issues (/agent-task)
      2. Run discovery to find work (/agent-baseline, /agent-review)
      3. Describe what you want to work on (I'll create an issue)