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.
git clone https://github.com/majiayu000/claude-skill-registry
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"
skills/data/agent-ops-focus-scan/SKILL.mdFocus 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
first for quick issue overview. Only load full priority files when selecting specific issues to work on.index.md
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
CLI installed. Focus scanning can be done via direct file reading.aoc
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)
- Read current
.agent/focus.md - Update "Doing now" section with selected issue
- 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:
| Command | Purpose |
|---|---|
| Find critical actionable issues |
| Find high priority issues |
| Find blocked issues |
| Check issue details |
| Quick status overview |
Confidence-Based Batch Limits
| Confidence | Max Issues per Iteration | Rationale |
|---|---|---|
| LOW | 1 (hard limit) | High uncertainty requires focused human oversight |
| NORMAL | Up to 3 | Standard workflow with reasonable batching |
| HIGH | Up to 5 | Well-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:
-
Check for existing issue: Is there an issue for the requested work?
- Yes → proceed to priority scan
- No → prompt to create issue first
-
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}" -
Even simple chores need issues:
- "Fix typo" →
issueCHORE - "Update dependency" →
issueCHORE - "Add comment" →
issueDOCS - "Quick refactor" →
issue (needs permission anyway)REFAC
- "Fix typo" →
Procedure
-
Check confidence level (from constitution or task):
- Determine batch size limit
- Note if LOW confidence (special handling)
-
Scan priority files in order (stop when actionable issue found):
(scan first — highest priority).agent/issues/critical.md.agent/issues/high.md.agent/issues/medium.md
(only if nothing higher).agent/issues/low.md- SKIP
— backlog items need triage first.agent/issues/backlog.md
-
Identify the highest priority issue where:
- Status is
ortodoin_progress - All
dependencies aredepends_on
(or list is empty)done - Issue is not marked as
,blocked
, orcancelleddropped
- Status is
-
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 differentFor 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 onlyFor HIGH confidence:
- Proceed with selection (up to 5), report only
-
Action based on findings:
If an actionable issue is found:
- Read
.agent/focus.md - Update
to reflect this issue (include issue ID)Doing now - Update
with the immediate practical stepNext - 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)
- Read