Agent Skills: Work Backlog Item

<provided_arguments> $ARGUMENTS </provided_arguments>

<provided_arguments/> is free-text — flags, positional values, and/or a freetext suffix — either typed by a human invoking /dh:work-backlog-item ... directly, or already known to you when you are initiating this action yourself (e.g. a "capture a backlog item" or "groom/work item #N" decision you made in this session). In the self-initiated case you already know every field below directly — go straight to producing the coerced result; there is nothing in <provided_arguments/> you don't already know, and no reason to route it through anything else.

Coerce <provided_arguments/> to match this schema yourself, using the vocabulary below, then treat the result as <input/>: parse.schema.json.

Argument vocabulary:

• Route — the first positional word only, when it matches one of the keys registered in command-routes.json: create, groom, work, close, resolve, setup-github, progress, resume. Because only the first positional is ever checked, at most one route keyword can ever be found in a single invocation — there is no such thing as a "two routes" conflict. The same word appearing later (e.g. inside a title) is not a route. No match on the first positional → route is title_substring (positionals or freetext remain) or none (nothing at all — no flags, no positionals, no freetext).
• item_ref discriminator — any positional matching #N, bare digits, or a GitHub issue URL (https://github.com/{owner}/{repo}/issues/N) → normalize to #N (keep a URL verbatim). Checked across all positionals, not just the first — including one embedded inside an otherwise-ordinary title (verified: Fix bug on line 42 → {"route":"issue","item_ref":"#42","user_text":"Fix bug on line"} — the "42" is consumed as item_ref and removed from user_text, and route becomes issue rather than title_substring; be alert to this when a title happens to end in a number). When it is the only discriminator found (no registry route word present), route is the literal string issue — not title_substring — and no reference key is set (verified: #42 alone → {"mode":"interactive","route":"issue","item_ref":"#42"}). A route word and one item_ref may both be present (e.g. groom #50, close #42) — that is route + item_ref together (registry route wins as route, reference is set per the route table above), not a conflict. Two or more item_ref discriminators in one invocation (e.g. close #42 #55) is the one real conflict case — ask the user to disambiguate rather than picking one.
• Freetext delimiter — --, or a bare —/– (em/en dash — a mobile-autocorrect artifact; normalize a leading —/– on any token to --). Everything after the delimiter is user_text verbatim, regardless of content (quotes, code, punctuation — do not further tokenize it). No delimiter → user_text is whatever positionals remain after removing the route/item_ref tokens, space-joined. Exception when flags.quick is present: do not apply this delimiter rule — positionals before a -- are otherwise discarded (captured nowhere, since they don't match item_ref's pattern), which would silently drop part of the supplied request. Still remove recognized flags and their values first (per the Flags rule below — e.g. --quick --auto Fix login strips --auto, not just its own name), and still apply the item_ref discriminator only when it consumes the entire remaining text (e.g. --quick #42 alone → item_ref="#42", nothing left for user_text) — a discriminator match embedded in a longer request (e.g. --quick #42 fails on SSO) does not fire, since the surrounding text is exactly what quick/start.md Step 1 needs to derive title/observations from. After flags and a whole-text-only item_ref are handled, set user_text to whatever remains verbatim (any embedded -- is literal text, not a delimiter). Step 1 derives its own title and observations from that raw text — it is not pre-split here.
• Flags — --language <value>, --stack <value>: both take the next token as their value, but only when that next token does not itself start with - — a next token starting with - (including no next token at all) means the value is missing, a stop-and-ask condition, not "consume the next flag as this flag's value" (verified: --language --stack python-fastapi treats --language as missing its value; it does not consume --stack as the value). --force, --auto, --quick (boolean, no value). mode is auto only when --auto is present, otherwise interactive.
• --help/-h present → show usage (this vocabulary plus argument-hint in the frontmatter) and stop; do not route.

Route → reference file: see command-routes.json — one JSON object, route keyword to reference-file path, do not hand-copy it here.

For every placeholder in the form <key/>, substitute the value of that key from <input/>.

<sam_cli> uv run "${CLAUDE_PLUGIN_ROOT}/sam_schema/cli.py" </sam_cli>

<mcp_server_scripts> SAM server: uv run --script "${CLAUDE_PLUGIN_ROOT}/scripts/run_sam_server.py" Backlog server: uv run --script "${CLAUDE_PLUGIN_ROOT}/scripts/run_backlog_server.py" --project-dir . </mcp_server_scripts>

The references/workflows/*.md files loaded by this skill are plain files, not substituted — they show bare SAM CLI subcommands and args only (e.g. backlog view --selector "..."), never the invocation prefix. Prepend the command in <sam_cli/> above to every one of them.

[!IMPORTANT] When provided a process map or Mermaid diagram, treat it as the authoritative procedure. Execute steps in the exact order shown, including branches, decision points, and stop conditions. A Mermaid process diagram is an executable instruction set. Follow it exactly as written: respect sequence, conditions, loops, parallel paths, and terminal states. Do not improvise, reorder, or skip steps. If any node is ambiguous or missing required detail, pause and ask a clarifying question before continuing. When interacting with a user, report before acting the interpreted path you will follow from the diagram, then execute.

The following diagram governs argument coercion:

flowchart TD
    %% No subprocess, no shell — coerce provided_arguments directly against the vocabulary above.
    Coerce["Coerce provided_arguments to the<br>parse.schema.json shape using the vocabulary above"] --> ReqCheck{"mode and route both derivable?"}
    ReqCheck -->|"No"| ErrReq(["STOP — ask for clarification;<br>mode/route are always derivable from the<br>vocabulary above, a miss here means the<br>input didn't match any covered shape"])
    ReqCheck -->|"Yes"| ConflictCheck{"two or more item_ref<br>discriminators present, or a flag<br>missing its required value?"}
    ConflictCheck -->|"Yes"| ErrConflict(["STOP — ask the user to disambiguate"])
    ConflictCheck -->|"No"| RouteCheck{"route is one of the registry<br>keywords (create/groom/work/close/<br>resolve/setup-github/progress/resume)?"}
    RouteCheck -->|"No — route is none, title_substring, or issue"| Ready(["input ready — proceed to routing"])
    RouteCheck -->|"Yes"| RefLookup["Set reference from<br>command-routes.json"]
    RefLookup --> NeedRefCheck{"route is close or resolve,<br>and item_ref/user_text both absent?"}
    NeedRefCheck -->|"Yes"| ErrNoTarget(["STOP — ask which item to<br>close/resolve; close/start.md's<br>Step 5.2 selector requires one"])
    NeedRefCheck -->|"No"| Ready

In auto mode, do not call AskUserQuestion. Log each would-be interactive decision as [AUTO] {decision} - {evidence}.

The following diagram governs pipeline stage execution:

flowchart TD
    %% Pipeline order: create -> groom -> work. Each earlier stage runs only if its output is missing.
    %% Only reached for route in {create, groom, work} — the route-dispatch diagram above sends
    %% `issue`/`title_substring` routes here as `work` (item_ref/user_text identifies the item);
    %% every other route (close/resolve/setup-github/progress/resume/none) is terminal there and
    %% never reaches this diagram.
    RouteIn(["route value from parsed JSON"]) --> RouteCheck{"route value?"}

    RouteCheck -->|"create"| CreateItemRef{"valid item_ref<br>already available?"}
    CreateItemRef -->|"Yes — existing ref from input<br>or GitHub issue URL"| CreateSkip(["STOP — item already exists, creation not needed"])
    CreateItemRef -->|"No — no existing ref"| CreateScope["Read scope.md<br>references/workflows/create/scope.md"]
    CreateScope --> RunCreate["Run create workflow<br>references/workflows/create/start.md"]
    RunCreate --> CreateDone{"item_ref now exists<br>in parsed state?"}
    CreateDone -->|"No — creation failed"| CreateFail(["STOP — report creation failure"])
    CreateDone -->|"Yes — item created"| CreateEnd(["STOP — creation complete"])

    RouteCheck -->|"groom"| GroomItemRef{"valid item_ref<br>already available?"}
    GroomItemRef -->|"No"| GroomCreate["Run create workflow first<br>references/workflows/create/start.md"]
    GroomCreate --> GroomStart
    GroomItemRef -->|"Yes"| GroomStart["Run groom workflow<br>references/workflows/groom/start.md"]
    GroomStart --> GroomDone(["STOP — grooming complete"])

    RouteCheck -->|"work"| WorkItemRef{"valid item_ref<br>already available?"}
    WorkItemRef -->|"No"| WorkCreate["Run create workflow first<br>references/workflows/create/start.md"]
    WorkCreate --> WorkGroomCheck
    WorkItemRef -->|"Yes"| WorkGroomCheck{"grooming already complete<br>for this item?"}
    WorkGroomCheck -->|"No — grooming incomplete"| WorkGroom["Run groom workflow<br>references/workflows/groom/start.md"]
    WorkGroom --> WorkGate
    WorkGroomCheck -->|"Yes — grooming confirmed complete"| WorkGate{"gate blocks progression?<br>prerequisites missing or item<br>explicitly marked BLOCKED?"}
    WorkGate -->|"Yes — gate blocked"| WorkBlocked(["STOP — report blocking reason<br>and missing prerequisites"])
    WorkGate -->|"No — gate clear"| WorkRun["Run work workflow<br>references/workflows/work/start.md"]
    WorkRun --> WorkEnd(["Work workflow complete"])

Work Backlog Item

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

See the Backlog Lifecycle reference for the complete state machine, handoff protocol, and data architecture.

Phase separation: Grooming (Step 3.1) is autonomous research — the agent verifies facts, maps resources, estimates effort, and surfaces blockers. Planning (Step 4.2) 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). The configured backend is authoritative for its native work records. For Beads-backed projects, use bd directly for issue creation, inspection, status, dependencies, readiness, labels, notes, and metadata. Use MCP or the provider-neutral CLI for structured plans, dispatch, artifacts, validation, and handoffs that Beads does not provide. Do not describe MCP or CLI as an exclusive proxy layer.

MCP server availability: Both plugin:dh:backlog and plugin:dh:sam initialize in ~1–2 seconds after a session restart. Claude Code handles connection waiting automatically. If a tool is unavailable, see the troubleshooting steps at ${CLAUDE_PLUGIN_ROOT}/docs/mcp-connection-check.md — its commands use the <sam_cli/> and <mcp_server_scripts/> values above.

To capture a new backlog item: /dh:work-backlog-item create -- "<what and why of the problem that triggered the need for a backlog issue>"

Speculative causes: if you state a guess at why something is broken, it's recorded as **Hypothesis**: {text} rather than fact — grooming later confirms or refutes it and rewrites that line to match what was actually found, so the item never keeps showing an unverified guess as settled.

Arguments

route is none only when argv is empty (no flags, no positionals, no freetext suffix): follow Step 1.1 — Interactive Browser below. It is not the same as mode: "interactive" (which only means --auto was not passed).

On backend=beads: a beads ID (bd-a3f8) coerces to title_substring/user_text, not item_ref — find_item() still resolves it via its string-ID exact-match branch, so this is a routing detail, not a functional gap.

Optional flags (when route is title_substring, issue, or a pipeline route): --language <lang> selects language plugin (default: python); --stack <profile> selects stack profile (e.g., python-fastapi, python-cli). See sdlc-layers.

/work-backlog-item                                    # interactive browser
/work-backlog-item #42                               # issue-first → planning
/work-backlog-item 42                                # issue-first (bare number) → planning
/work-backlog-item https://github.com/{OWNER}/{REPO}/issues/42  # URL → 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              # dismiss by title
/work-backlog-item close #42                         # dismiss by issue number
/work-backlog-item resolve Error Recovery            # mark completed by title
/work-backlog-item resolve #42                       # mark completed by issue number
/work-backlog-item --language python --stack python-fastapi Add auth  # Layer 2 stack profile

--quick mode

Loads references/workflows/quick/start.md with flags.quick = true in the coerced input and item_ref set to the raw request (e.g. #N for an existing issue). For one-file fixes where full grooming is disproportionate. The request can be a question, a report, or a fix ask — it does not need to already read like a title. The agent derives a short title plus its own observations about what's being asked from that raw text, per Step 1; it does not just mirror the raw text into both fields:

/work-backlog-item --quick why does login keep redirect-looping when SSO is enabled

Proactive fix routing: The Proactive Fix Gate in .claude/CLAUDE.md (Proactive Fix Gate section) routes trivial discovered issues to this --quick path autonomously.

--auto mode rules

All interactive AskUserQuestion calls are replaced with evidence-derived decisions. Load auto-mode.md for the full substitution table. BLOCKED states (RT-ICA MISSING conditions, feasibility gate BLOCKED) require human resolution regardless of mode.

Workflow

Routing (evaluated first, before any step)

The following diagram governs route dispatch:

flowchart TD
    %% Dispatch runs once <input/> is ready. flags.quick takes priority over route.
    %% Dispatch is entirely on the route field — never on flags.auto/flags.quick, which
    %% are independent boolean modifiers, not route values (verified: `--auto` alone
    %% produces route="title_substring", flags={auto:true} — never route="auto").
    Start(["input ready"]) --> QuickCheck{"flags.quick present?"}
    QuickCheck -->|"Yes"| SQ["Load references/workflows/quick/start.md<br>item_ref = item_ref field, or the quick-mode user_text<br>(raw text after --quick, unsplit) as the request to derive title/observations from"]
    SQ --> SQEnd(["STOP — quick workflow handles session"])
    QuickCheck -->|"No"| Q1{"route value?"}

    Q1 -->|"none"| S0["Load references/workflows/work/interactive-browser.md<br>Step 1.1 — interactive browser"]
    S0 --> S0End(["STOP — interactive browser handles session"])

    Q1 -->|"create"| PipelineCreate(["Continue to pipeline stage execution<br>(route = create)"])
    Q1 -->|"groom"| PipelineGroom(["Continue to pipeline stage execution<br>(route = groom)"])
    Q1 -->|"work"| PipelineWork(["Continue to pipeline stage execution<br>(route = work)"])
    Q1 -->|"issue or title_substring"| PipelineEntry(["Continue to pipeline stage execution<br>as if route = work — item_ref or user_text<br>identifies the item; its own<br>create-if-missing / groom-if-incomplete<br>chain handles the rest"])

    Q1 -->|"progress"| SP["Load references/workflows/progress/start.md<br>item_ref = item_ref field, if present"]
    SP --> SPEnd(["STOP — progress report handles session"])

    Q1 -->|"resume"| SR["Load references/workflows/resume/start.md<br>item_ref = item_ref field, if present"]
    SR --> SREnd(["STOP — resume workflow handles session"])

    Q1 -->|"close"| S9c["Load references/workflows/close/start.md<br>item_ref = item_ref field, or user_text as title"]
    S9c --> S9cEnd(["STOP — close workflow handles session"])

    Q1 -->|"resolve"| S9r["Load references/workflows/close/start.md<br>item_ref = item_ref field, or user_text as title"]
    S9r --> S9rEnd(["STOP — resolve workflow handles session"])

    Q1 -->|"setup-github"| SGH["Load references/workflows/setup-github/start.md"]
    SGH --> SGHEnd(["STOP — setup-github workflow handles session"])