List Work Items

!accelerator config context --skill list-work-items --fail-safe !accelerator config agents --fail-safe

If no "Agent Names" section appears above, use these defaults: accelerator:reviewer, accelerator:codebase-locator, accelerator:codebase-analyser, accelerator:codebase-pattern-finder, accelerator:documents-locator, accelerator:documents-analyser, accelerator:web-search-researcher.

Work items directory: !accelerator config path work --fail-safe Work item ID pattern: !accelerator config work id_pattern --fail-safe Local ID prefix: !accelerator config work key --fail-safe Active integration: !accelerator config work integration --fail-safe

The reads above inform how you parse the filter argument. accelerator work list owns the sync-status rendering itself: it adds the Sync column only when a last-sync.json baseline exists for the configured work.integration or some listed item is a draft, and degrades to presence-only when the remote is unreachable — you do not branch on the integration here.

Work Item Template

The following template defines the work item schema and field defaults.

!accelerator config template work-item --fail-safe

Known kind values: !accelerator work template-hints kind Known status values: !accelerator work template-hints status Known priority values: !accelerator work template-hints priority

Each line above lists one value per line. Collect these into three sets: known kinds, known statuses, and known priorities. These hints inform the filter shorthand rules below but do not restrict what values may appear on work items — legacy values like todo, done, or adr-creation-task are matchable only via the explicit structured form (rule 2).

You are tasked with listing and filtering work items from the configured work items directory. This is a read-only skill — never write any files and never spawn sub-agents. Your job is to parse the user's natural-language filter into accelerator work list's flags and present its output; the CLI owns the scan, filter, sync classification, and table/hierarchy rendering.

Step 1: Resolve Filter

If an argument was provided, parse it as a filter expression using the following precedence rules (rules 3–4 use the known kind/status/priority sets collected above). The first rule that matches wins.

Filter Precedence Rules

Rule 1 — Presentation keywords (changes rendering, not filtering): hierarchy, as a tree, show hierarchy. These may combine with a filter (e.g. hierarchy under 0042). Strip the keyword from the argument and continue parsing the remainder, if any, through rules 2–5.

Rule 2 — Explicit structured forms (keyword identifies the field):

• tagged <value> or with tag <value> → filter by tag
• under <value> or children of <value> → filter by parent
• status <value> → filter by status (matches any value on any work item)
• kind <value> → filter by kind (matches any value on any work item)
• priority <value> → filter by priority
• about <text> → free-text title search (case-insensitive substring)

Rule 3 — Multi-token template-value shorthand: two or more tokens that each match a known template-comment value in different fields. For example, bugs in review → kind: bug AND status: review (after singularising bugs to bug and recognising in-progress, review, etc. as status values with filler words like in, only, all stripped). Both tokens must match values from distinct fields. If either token is ambiguous across fields, fall through to rule 5.

Rule 4 — Single-token template-value shorthand: one token matching a known kind, status, or priority value from the template comments. Map common plurals (bugs→bug, epics→epic, stories→story, tasks→task, spikes→spike) and common synonyms (drafts→draft). If the token matches values in more than one field, ask the user for disambiguation rather than guessing.

Rule 5 — Free-text title search: anything that does not match rules 1–4 is treated as a case-insensitive substring search against the title: frontmatter field.

If no argument was provided: no filter — list every work item.

Translate the parsed filter into work list flags

Once the argument is parsed, map each recognised clause onto a concrete flag. This translation is the only work the skill does; the CLI applies the filters and echoes the interpreted filter and match count itself, so do not echo it separately.

A multi-token shorthand (rule 3, e.g. bugs in review) becomes several flags (--kind bug --status review). Pass the canonical value the parse resolved (singularised, synonym-mapped); --parent is canonicalised by the CLI, so pass the bare form the user gave. Every flag is a conjunct — an item must satisfy all of them.

Step 2: Run work list and present its output

Invoke the CLI with the translated flags:

accelerator work list \
  [--status <s>] [--kind <k>] [--priority <p>] [--parent <ref>] \
  [--tag <t>]... [--hierarchy] [<title-substring>]

The CLI does everything the previous scan/filter/render steps did, so there is nothing to reconstruct in prose:

• Scan and validity. It reads every *.md file in the work directory and in its drafts/ subdirectory, treats a file as a work item only when it has closed frontmatter and a non-empty id (or legacy work_item_id), silently excludes non-items, and warns "<filename>: skipped — no frontmatter" / "… unclosed frontmatter" on malformed files. The frontmatter id is the displayed ID.
• Drafts. A draft item lives in drafts/ under a provisional draft-xxxxxx ID because no tracker has confirmed it yet. Being a draft item is unrelated to status: draft: a draft item can carry any status, and a status: draft item need not be a draft item.
• Filter. Each flag is applied as a conjunct; --parent canonicalises both sides (short and long ID forms compare equal); the title term is a case-insensitive substring. The CLI prints the interpreted filter and match count itself (Filter: status=draft (3 matches), Children of 0042 (2 matches), or All work items (29 total)).
• Sync column. It appends a Sync column when a last-sync.json baseline exists for the configured work.integration or any listed item is a draft, and omits it otherwise. A draft item always reads 🟠 draft. With a baseline, every other item carries the five-state label vocabulary (🟢 synced, ⚪ unsynced, 🔵 locally modified, 🟣 remotely modified, 🔴 conflict), driven by one bulk remote read through the shared classifier; without one it carries the presence-only label (🟢 synced when it has an external_id, else ⚪ unsynced). If the remote is unreachable it degrades to presence-only and still exits 0 — never retrying or hanging.

• Hierarchy. --hierarchy renders the parent/child tree with Unicode box-drawing characters, appends each line's sync label, marks an out-of-set parent (parent … not found), and detects cycles so rendering always terminates.

  The parent/child tree renders as, for example:

NNNN — parent title (kind: <kind>, status: <status>) ├── NNNN — child 1 title (kind: <kind>, status: <status>) ├── NNNN — child 2 title (kind: <kind>, status: <status>) └── NNNN — last child title (kind: <kind>, status: <status>)

• Empty and missing. A directory that does not exist prints the paths.work guidance and exits 0; an empty directory prints No work items found in …; a filter that matches nothing prints No work items matched: … rather than an empty table.

Print the CLI's stdout to the user as-is (it is markdown-native — a table or tree the conversation renders directly), and surface any stderr warnings it emitted. Do not re-echo the filter or re-render the table.

Quality Guidelines

• Read-only: never write any files. This skill only reads and displays.
• No sub-agents: never spawn sub-agents. All work is a natural-language filter parse plus one accelerator work list invocation.
• No hardcoded field values: never assume a specific set of status, kind, or priority values. The template's comments list shipping defaults, not a closed set. Users may override the template with custom values.
• Explicit structured filters are universal: status <value>, kind <value>, etc. match any value present on any work item, not just template defaults. This is how legacy values like todo or adr-creation-task are reachable.
• Resilient to malformed work items: missing or unclosed frontmatter must not crash the listing — warn using the resolved filename and continue. Warning phrasing should match work show's own errors: "no frontmatter" / "unclosed frontmatter".
• Frontmatter id is authoritative: the id field (or work_item_id on legacy files) is the work item ID, even if the filename prefix differs. This applies to legacy bare-number, project-coded, tracker-keyed and draft items alike.
• Hierarchy safety: hierarchy rendering must terminate in bounded time even if parent cycles exist. Detect cycles and render affected work items flat with a marker.

!accelerator config instructions list-work-items --fail-safe