Work Backlog Item

Work Backlog Item

Bridge a backlog item into the SAM planning pipeline via /python3-development:add-new-feature (default). Optional --language and --stack select Layer 1/2 profiles — see .claude/docs/sdlc-layers/.

Phase separation: Grooming (Step 3) is autonomous research — the agent verifies facts, maps resources, estimates effort, and surfaces blockers. Planning (Step 6) is solution design — architecture, tasks, implementation. The human sets priorities and resolves blockers; the agent handles research and fact-checking autonomously.

SAM — Stateless Agent Methodology. See sam-definition.md for what SAM is and how to embody it. SAM lives in ../stateless-agent-methodology/ (or bitflight-devops/stateless-agent-methodology on GitHub). Primary source of truth is GitHub Issues (labels + milestone = canonical status); .claude/backlog/ per-item files are the local cache and are kept in sync.

When invoked with no arguments, shows an interactive browser. When invoked with #N or a title substring, proceeds directly to the planning workflow.

Arguments

$0 selects the operating mode; remaining positional args ($1, $2, ...) form the title or parameter:

Optional flags (when $0 is title substring or --auto): --language <lang> selects language plugin (default: python); --stack <profile> selects stack profile (e.g., python-fastapi, python-cli). See .claude/docs/sdlc-layers/.

/work-backlog-item                                    # interactive browser
/work-backlog-item #42                               # issue-first → planning
/work-backlog-item Error Recovery                    # direct match → planning
/work-backlog-item --auto                            # autonomous → auto-select first open P0/P1
/work-backlog-item --auto vercel skills npm package  # autonomous → planning
/work-backlog-item close Error Recovery              # verify and close by title
/work-backlog-item close #42                         # verify and close by issue number
/work-backlog-item resolve commitlint                # mark no longer applicable
/work-backlog-item resolve #17                       # mark no longer applicable by issue
/work-backlog-item --language python --stack python-fastapi Add auth  # Layer 2 stack profile

--auto mode rules

When $0 is --auto, the following substitutions apply at every interactive decision point:

--auto does NOT change the behaviour of Steps 3–8 (grooming, RT-ICA evaluation, SAM planning, backlog update) — those are already agent-executable without human input.

Workflow

Routing (evaluated first, before any step)

Dispatch based on $0 (the first argument word) before executing any step:

AUTO_MODE — when set, all AskUserQuestion calls are replaced with evidence-derived decisions. See the --auto mode rules table in the Arguments section for each substitution.

Step 0: Interactive Browser (no arguments only)

Trigger: $0 is empty (no arguments passed).

1. Invoke the backlog script to list items with status:

   uv run .claude/skills/backlog/scripts/backlog.py list --format json --with-status -R Jamie-BitFlight/claude_skills
   

   Parse the JSON output. Each entry has section, title, issue, plan, status, milestone, file_path (index format), groomed (true if item has groomed content).

2. Groomed = item has groomed: true in JSON, or ## Groomed section in its per-item file (.claude/backlog/{priority}-{slug}.md). Read the item file; if groomed sections present, use them.

3. Present a numbered list. Use these status indicators in user-visible output only:

   Backlog Items:
   
   P0
     1. ✅ SAM: Error Recovery / Rollback Procedures       [#12  status:in-progress  v1.0]
     2. 🔍 SAM: Regex False Positive Suppression           [#14  status:needs-grooming  v1.0]
   
   P1
     3. 📋 SAM: Validate Task File Schema                  [no issue]
     4. 📋 SAM: Implement Feature Dry-Run Mode             [no issue]
   
   P2
     5. 🔍 SAM: Context Window Budget Tracking             [#18  status:needs-grooming  —]
   
   Ideas
     6. 📋 SAM: Multi-Repo Support                         [no issue]
   
   Status: ✅ = planned/in-progress  🔍 = groomed/needs-grooming  📋 = not yet groomed
   
   Options:
     [number]   — Select item to work on
     G [number] — Groom a specific item
     G all      — Groom all ungroomed items
     D [number] — Show full details for an item
   

4. Use AskUserQuestion to ask: "Which item would you like to work on next?"

5. Handle the response:

   • [number] — use that item's title as the working title and proceed to Step 1
   • G [number] — invoke Skill(skill="groom-backlog-item", args="{item title}") then re-display the list
   • G all — invoke Skill(skill="groom-backlog-item", args="all") then re-display the list
   • D [number] — display the full item description, research_first field, and groomed content (if present in the item file under ## Groomed), then re-display the list
   • C [number] — proceed to Step 9 (close path) with that item's title
   • R [number] — proceed to Step 9 (resolve path) with that item's title

Routing: If $0 is close or resolve, extract $1+ as the title and jump directly to Step 9.

Step 1b: Issue-First Path (#N)

Trigger: $0 matches #[0-9]+.

Extract the issue number {issue_number} from $0. Fetch the issue:

gh issue view {issue_number} -R Jamie-BitFlight/claude_skills \
  --json number,title,state,body,labels,milestone

If the issue does not exist, report and stop. If the issue is already closed, warn: "Issue #{issue_number} is closed. Use close or resolve if needed." and stop.

From the issue response build the working item:

Then try to find a matching per-item file in .claude/backlog/ by issue number (metadata.issue) or title. If found, use it to supplement any missing fields (e.g. research_first, suggested_location). If not found, continue without a local per-item file — the GitHub Issue is sufficient.

Proceed to Step 2.7 (Set In-Progress Label) with the assembled item, then continue to Step 3.

Step 1: Find the Backlog Item

Run uv run .claude/skills/backlog/scripts/backlog.py list --format json and search the title field of each entry for a case-insensitive match against the title. Title = $1+ joined (args after the mode flag $0). In interactive mode, title = full $ARGUMENTS.

AUTO_MODE with no title ($1 is empty): apply the "No title given" substitution from the --auto mode rules table — scan P0 then P1 sections for the first open item, log and use its title. Skip items with status: done or status: resolved in their JSON entry (these are filtered out by backlog.py list already).

• Zero matches (interactive mode): report "No backlog item found matching: {title}" and offer to create one via /create-backlog-item.
• Zero matches (AUTO_MODE): log [AUTO] No item found — invoking create-backlog-item --auto {title}, invoke Skill(skill: "create-backlog-item", args: "--auto {title}"), then re-run Step 1.
• Multiple matches (interactive mode): list all matches and ask the user to pick one.
• Multiple matches (AUTO_MODE): log [AUTO] Multiple matches — picking first: {title}, proceed with first match.

Record the priority section (P0, P1, P2, Ideas) the item belongs to.

Step 2: Extract Item Fields

From the matched item's JSON output (via backlog.py list --format json), extract title, plan, section (priority), issue, groomed, and file_path. For detailed fields not in JSON (description, source, added, research_first, suggested_location), read the per-item file at file_path.

• title — the title field from JSON (required)
• source — not in JSON; read from per-item file frontmatter metadata.source if needed (optional)
• added — not in JSON; read from per-item file frontmatter metadata.added if needed (optional)
• description — not in JSON; read from per-item file frontmatter description (required)
• research_first — not in JSON; read from per-item file body **Research first**: line (optional)
• suggested_location — not in JSON; read from per-item file body **Suggested location**: line (optional)
• plan — the plan field from JSON (optional)

If the item already has a **Plan**: field, report:

This item already has a plan at {path}. Use /python3-development:implement-feature {path} to execute it.

Then stop.

After extracting fields, proceed to Step 2.5 (GitHub Issue Sync) before continuing to Step 3.

Step 3: Auto-Groom (if needed)

1. Check if item is groomed: Check the groomed field in the JSON output. If true, read the per-item file at file_path for the groomed content under ## Groomed. Use that content.
2. Search conversation context for a recent groom-backlog-item output matching this item.

If no groomed content exists in the item file:

Skill(skill: "groom-backlog-item", args: "{item title}")

The groom skill writes groomed content into the per-item file. Capture the groomed output (Reproducibility, Resources, Dependencies, Blockers, etc.) for use in the feature request.

Step 4: RT-ICA Checkpoint

1. Goal statement — What completing this item achieves.
2. Reverse prerequisites — Conditions required for success (enumerate each).
3. Availability check — For each condition: AVAILABLE / DERIVABLE / MISSING.
4. Decision — APPROVED or BLOCKED.

If BLOCKED:

When RT-ICA blocks, optionally offer ARL human-probing questions (e.g., "What went wrong in the past?", "What references are essential?") to capture invisible knowledge. Add answers to .claude/domain-knowledge/ with staleness tracking. See .claude/docs/sdlc-layers/arl-human-probing-design.md.

Present a structured summary to the user:

RT-ICA: BLOCKED

Missing inputs that prevent SAM planning:
- {missing condition 1}
- {missing condition 2}

Provide these inputs before proceeding. SAM planning will not be invoked with known gaps.

Wait for user response. Do not invoke Step 5 until BLOCKED is resolved.

If APPROVED:

Proceed to Step 5. Carry DERIVABLE items forward as "Assumptions to confirm" in the RT-ICA section of the feature request.

Step 5: Compose Feature Request

Build the feature request string for add-new-feature. If --stack was specified, append a "Stack profile" line to the feature request. If --language was specified and is not python, invoke the corresponding language plugin (e.g., /typescript-development:add-new-feature for typescript).

## Backlog Item: {title}

**Source**: {source}
**Priority**: {priority section — P0/P1/P2/Ideas}
**Added**: {added date}

### Description

{description text}

### Research Questions

{research_first text, or "None" if absent}

### Suggested Location

{suggested_location text, or "To be determined during architecture phase" if absent}

### RT-ICA Assessment

**Decision**: APPROVED
**Goal**: {goal statement}
**Verified conditions**: {list of AVAILABLE items}
**Assumptions to confirm**: {list of DERIVABLE items, or "None"}

### Grooming Context

{full context manifest from Step 3, if available}

### Stack Profile (optional)

{stack profile name if --stack specified, e.g., python-fastapi}

Step 6: Invoke SAM Planning

Skill(skill: "python3-development:add-new-feature", args: "{composed feature request}")

This runs the full SAM workflow: discovery, codebase analysis, architecture spec, task decomposition, validation, context manifest.

Step 7: Update Backlog with Plan Reference

After add-new-feature completes, identify the task file it created:

Glob(pattern="plan/tasks-*-{slug}*")

Where {slug} is the item title lowercased with spaces replaced by hyphens.

Invoke the backlog script to add the Plan:

uv run .claude/skills/backlog/scripts/backlog.py update "{title}" --plan "plan/tasks-{N}-{slug}.md" -R Jamie-BitFlight/claude_skills

If the item has **Issue**: #N, record it in the plan file header comment and include Fixes #N in any commit message produced during implementation.

Step 8: Report Next Steps

Backlog item "{title}" is now planned.

- Plan file: plan/tasks-{N}-{slug}.md (or plan/tasks-{N}-{slug}/ directory)
- To execute:      /python3-development:implement-feature {slug}
- To check status: /python3-development:implementation-manager status . {slug}
- To close when done: /work-backlog-item close {slug}

Do NOT close the GitHub Issue directly. Include Fixes #N in commit messages and the PR body — the issue auto-closes when the PR merges. Only use /work-backlog-item close for post-merge verification and local bookkeeping. Never invoke backlog.py close before the PR has merged.

Step 9: Verify and Close

Trigger: $0 is close or resolve.

Extract the operation from $0 and the argument from $1+:

• $0 = close: $1+ = title or #N → verify implementation and mark COMPLETED
• $0 = resolve: $1+ = title or #N → mark no longer applicable (no verification required)

9a: Find Item

If $1 starts with # (e.g., close #42), treat it as an issue number:

gh issue view {issue_number} -R Jamie-BitFlight/claude_skills \
  --json number,title,state,body,labels

• If the issue is not found, report and stop.
• Extract title from the issue response and use it as the working title.

Otherwise, scan .claude/backlog/ per-item files and search item titles for case-insensitive match against {title}.

• Zero matches: report "No backlog item found matching: {title}" and stop.
• Multiple matches: list all matches and ask user to pick one.
• Item already in ## Completed section: report "Item already closed on {Completed date}" and stop.

9b: Resolve path (skip verification)

If operation is resolve:

1. Use AskUserQuestion to ask: "Why is this item no longer applicable?" (free text)
2. Invoke the backlog script:

uv run .claude/skills/backlog/scripts/backlog.py resolve "{title or #N}" --reason "{reason}" -R Jamie-BitFlight/claude_skills

3. Report the script output to the user.

Then stop.

9c: Close path — checklist verification

If operation is close:

1. Extract **Plan**: field from the matched item. If absent:

No plan file recorded for "{title}". Cannot verify checklist.
Either run /work-backlog-item {title} first to create a plan,
or use /work-backlog-item resolve {title} if no plan was needed.

Then stop.

2. Read the plan file. Count:

   • total_tasks — lines matching - \[ \] or - \[x\]
   • checked_tasks — lines matching - \[x\]

3. If checked_tasks < total_tasks:

Checklist incomplete: {checked_tasks}/{total_tasks} tasks done.

Remaining:
{list of unchecked task lines}

Complete all tasks before closing this item.

Then stop.

9d: Close path — acceptance criteria verification

4. Spawn a verification agent:

Task(
  subagent_type: "general-purpose",
  prompt: "You are verifying whether a completed backlog item genuinely satisfies its stated goal.

Backlog item title: {title}
Description and acceptance criteria:
{description text from per-item file}

Plan file: {plan file path}
Plan checklist: {checked_tasks}/{total_tasks} — 100% complete.

Your task:
1. Read the plan file to understand what was implemented.
2. Search git log for commits referencing this item (use: git log --oneline -20).
3. Read 2-3 key changed files to verify the implementation exists.
4. Assess: Does the implementation satisfy the stated goal? Is the product better for it?

Return:
- PASS or FAIL
- One sentence of evidence (file:line or commit SHA)
- Any gaps you found (if FAIL)"
)

5. Collect agent verdict:
   • PASS: proceed to 9e
   • FAIL: report gaps, do not close:

Verification FAILED for "{title}".

Gaps found:
{agent findings}

Address these gaps before closing.

Then stop.

9e: Check for open PR

6. If the item has a linked GitHub Issue (#N), check whether an open PR already references it:

gh pr list -R Jamie-BitFlight/claude_skills --search "Fixes #N" --state open --json number,title,url

• Open PR found: The PR body contains Fixes #N — the issue will auto-close on merge. Update only the local per-item file status (do NOT close the GitHub Issue):

uv run .claude/skills/backlog/scripts/backlog.py update "{title}" --status in-progress -R Jamie-BitFlight/claude_skills

Report:

Backlog item "{title}" verified. GitHub Issue #{N} will auto-close when PR #{pr_number} merges.

Then stop.

• No open PR / no linked issue: proceed to 9f.

9f: Invoke backlog close

7. Invoke the backlog script (script updates per-item file and closes GitHub issue):

uv run .claude/skills/backlog/scripts/backlog.py close "{title}" --plan "{plan file path}" --checklist-pass -R Jamie-BitFlight/claude_skills

If invoked as close #N, use #N as the selector:

uv run .claude/skills/backlog/scripts/backlog.py close "#{N}" --plan "{plan file path}" --checklist-pass -R Jamie-BitFlight/claude_skills

8. Report the script output to the user.

GitHub Integration

.claude/backlog/ per-item files are the local cache. GitHub Issues are the source of truth. They are linked — when you work a P0/P1 backlog item, a GitHub Issue is created and kept in sync.

.claude/backlog/    →  GitHub Issue
  metadata.priority →  priority:* label
  Description       →  Issue body (story format)
  metadata.status   →  status:* label
  metadata.plan     →  Issue body Notes
  metadata.issue    ←  written back after creation
  metadata.status   →  Issue closed

Full step-by-step commands and example sessions: github-integration.md

Commit Messages and PR Body — Fixes #N

When implementing a backlog item that has a linked GitHub Issue (**Issue**: #N), every commit and the PR description MUST reference the issue so GitHub auto-closes it on merge:

• Commit messages: end the subject line with (fixes #N) or add a Fixes #N footer in the body:

  fix(perl-development): replace shell injection in run_command (fixes #80)
  

  or

  fix(perl-development): replace shell injection in run_command
  
  Fixes #80
  

• PR description: include at least one Fixes #N line in the PR body so GitHub links and auto-closes the issue when the PR is merged.
• When a PR fixes multiple issues, list each on a separate line:

  Fixes #80
  Fixes #83
  

• Use Fixes (not Closes) for bugs and security issues; either works for feature work.

This convention ensures issues are automatically closed on merge without manual intervention.

Step 2.5: GitHub Issue Sync

After Step 2, check for **Issue**: #N field in the matched item.

• Found: verify issue state with gh issue view {issue_number} -R Jamie-BitFlight/claude_skills --json number,title,state,labels
• Not found + P0/P1: offer to create a GitHub Issue (proceed to Step 2.5a)
• Not found + P2/Ideas: skip silently

Step 2.5a: Create GitHub Issue

Invoke the backlog script:

uv run .claude/skills/backlog/scripts/backlog.py update "{title}" --create-issue -R Jamie-BitFlight/claude_skills

The script creates the issue and writes issue: '#N' back to the per-item file frontmatter.

Step 2.7: Set In-Progress Label

If the item has **Issue**: #N, invoke the backlog script:

uv run .claude/skills/backlog/scripts/backlog.py update "{title}" --status in-progress -R Jamie-BitFlight/claude_skills

If the item is in a milestone with other issues, also run milestone start for the milestone:

uv run .claude/skills/gh/scripts/github_project_setup.py milestone start \
  --number {milestone_number} --repo Jamie-BitFlight/claude_skills

setup-github Command

Trigger: $0 is setup-github. Initializes label taxonomy, first milestone, and GitHub Project.

uv run .claude/skills/gh/scripts/github_project_setup.py labels --repo Jamie-BitFlight/claude_skills
gh api repos/Jamie-BitFlight/claude_skills/milestones -X POST \
  -f title="v1.0 — Skills Foundation" -f due_on="2026-03-31T00:00:00Z"
gh project create --owner Jamie-BitFlight --title "claude_skills Backlog"

Full setup steps and expected output: github-integration.md

Error Handling

• #N not found: report and list open issues with gh issue list -R Jamie-BitFlight/claude_skills --state open
• #N already closed: warn and stop; offer close or resolve if needed
• close #N / resolve #N — issue not found: report and stop
• Item not found: list available items from .claude/backlog/ per-item files with their priority sections
• Multiple matches: present numbered list, ask user to choose
• Grooming fails: proceed without grooming context, note the gap in the feature request
• RT-ICA returns BLOCKED: present missing inputs, wait for user, do not invoke add-new-feature
• add-new-feature fails: report the failure, do not update per-item file
• Plan file not found after planning: search plan/ directory broadly, ask user to confirm the path
• Grooming reports directory does not exist: treat all items as ungroomed
• close with no **Plan**: field: report and offer resolve as alternative
• close with incomplete checklist: list remaining tasks, do not close
• close with verification FAIL: report gaps, do not close
• close on already-completed item: report closed date, do not re-close
• resolve with no reason provided: block until user provides reason (reason is required evidence)
• GitHub issue creation fails: report error, continue with per-item-file-only workflow; do not block SAM planning
• gh not installed: run uv run .claude/skills/gh/scripts/setup_gh.py first
• Label not found during issue create: github_project_setup.py creates it automatically
• Milestone not found: skip milestone assignment; do not fail

Example Sessions

Issue-first planning (#N)

> /work-backlog-item #131

Loading GitHub Issue #131...
  Title:     plugin-validator: UX and coverage gaps
  Labels:    priority:p1, status:needs-grooming
  Milestone: v1.1 — Quality Gates
  State:     open

Matched per-item file for additional context: ✓
No groomed content in item file. Running groom-backlog-item first...

[groomed content written to item file]

RT-ICA: APPROVED — all conditions available.
Setting status:in-progress on issue #131...
  ✓ status:needs-grooming → status:in-progress

Composing feature request...
Invoking /python3-development:add-new-feature...

[SAM phases run]

Updated per-item file with Plan: plan/tasks-2-validator-ux-coverage.md

Next steps:
- To execute:      /python3-development:implement-feature validator-ux-coverage
- To close when done: /work-backlog-item close plugin-validator UX and coverage gaps

Planning (with title substring)

> /work-backlog-item Error Recovery

Found: "SAM: Error Recovery / Rollback Procedures" (P1)
No groomed content in item file. Running groom-backlog-item first...

[groomed content written to item file]

RT-ICA: APPROVED — all conditions available.
Composing feature request...
Invoking /python3-development:add-new-feature...

[SAM phases run]

Updated per-item file with Plan: plan/tasks-2-error-recovery.md

Next steps:
- To execute:      /python3-development:implement-feature error-recovery
- To check status: /python3-development:implementation-manager status . error-recovery
- To close when done: /work-backlog-item close error-recovery

Closing a completed item (by title)

> /work-backlog-item close validator UX

Found: "plugin-validator UX and coverage gaps" (P1)
Plan: plan/tasks-2-validator-ux-coverage.md
Checklist: 12/12 tasks complete

Spawning acceptance criteria verification agent...

Verdict: PASS
Evidence: Sub-issues 1-4 implemented in plugins/plugin-creator/scripts/plugin_validator.py
          commit 4a2f1b3 — "fix(validator): report unique files, add hook validation"

Backlog item "plugin-validator UX and coverage gaps" closed.
- Checklist: 12/12 tasks complete
- Acceptance criteria: PASS
- Status written to per-item file

Closing a completed item (by issue number)

> /work-backlog-item close #131

Fetching GitHub Issue #131...
  Title: plugin-validator UX and coverage gaps
  State: open

Found per-item file match: "plugin-validator UX and coverage gaps" (P1)
Plan: plan/tasks-2-validator-ux-coverage.md
Checklist: 12/12 tasks complete

Spawning acceptance criteria verification agent...

Verdict: PASS
Evidence: commit 4a2f1b3 — "fix(validator): report unique files, add hook validation"

Backlog item "plugin-validator UX and coverage gaps" closed.
- Checklist: 12/12 tasks complete
- Acceptance criteria: PASS
- Status written to per-item file
- GitHub Issue #131 closed

Resolving a no-longer-applicable item

> /work-backlog-item resolve commitlint verify last flag

Found: "commitlint: Verify --last flag and exit codes against primary sources" (P1)
Why is this item no longer applicable?
> REFUTED by fact-check: --last flag is verified in commitlint source cli.ts. No fix needed.

Backlog item resolved.
  Resolved: 2026-02-21
  Status: RESOLVED — REFUTED by fact-check: --last flag verified against commitlint
          source cli.ts and official docs. No fix needed.

GitHub-specific example sessions (issue creation flow and setup-github): github-integration.md

Validation Plan

See validation-plan.md for V1–V6 verification commands and the full integration test sequence.