/next-actions — Find Next Actions from KB Database

CRITICAL EXECUTION RULE: Do NOT spawn sub-agents (Task tool) for this skill. Call all MCP tools directly in the main conversation using the mcp__knowledge-base__ prefix. Sub-agents do not have reliable access to the knowledge-base MCP server and will waste time exploring the filesystem instead of making tool calls.

Usage

/next-actions [N] [--tag=TAG] [--plan-status=STATUS]

Examples:

# Show the next 5 actionable items (default)
/next-actions

# Show the next 3 actionable items
/next-actions 3

# Show the next 10 actionable items
/next-actions 10

# Show next actions for plans that improve testing
/next-actions --tag=testing

# Show next actions for lego-ui plans only
/next-actions --tag=lego-ui

# Show next actions from in-progress plans
/next-actions --plan-status=in-progress

# Combine: next 3 elaboration items from in-progress plans
/next-actions 3 --tag=elaboration --plan-status=in-progress

What It Does

Queries the KB database (single source of truth) for stories that need action, resolves dependencies at query time, and maps each story's state to the appropriate workflow command.

Inflight work always appears before new work. Stories being actively worked by an agent (active worktree in DB) are excluded entirely.

Parameters

Parameter	Required	Description
`N`	No	Number of items to return (default: 5)
`--tag=TAG`	No	Filter to stories linked to plans with this tag (e.g., `testing`, `lego-ui`, `elaboration`, `development`, `agent-tooling`)
`--plan-status=STATUS`	No	Filter to stories linked to plans with this status (`draft`, `accepted`, `stories-created`, `in-progress`, `implemented`, `superseded`, `archived`)

Priority Order (non-negotiable)

Rank	Category	States	Rationale
1	Needs QA	`ready_for_qa`, `in_qa`	Closest to done — unblock the pipeline first
2	Needs Fix	`failed_qa`, `failed_code_review`, `in_review`	Something was rejected — fix before starting new work
3	Needs Code Review	`ready_for_review`	Dev complete, waiting on review
4	Needs Dev	`ready`, `in_progress` (no active worktree)	Elaborated and ready to implement
5	Needs Elaboration	`backlog` (story file exists)	Story seed exists, not yet elaborated
6	Needs Story Creation	`backlog` (no story file)	Only a seed — story not yet generated

Within each rank, sort by priority: critical > high > medium > low.

Execution Steps

Step 1 — Get Active Worktrees

Call mcp__knowledge-base__worktree_list_active directly (no sub-agent):

mcp__knowledge-base__worktree_list_active({})
→ extract list of story_ids where status = 'active' → call these ACTIVE_IDS

If the tool is unavailable, set ACTIVE_IDS = [].

Step 2 — Query the KB Database

Call mcp__knowledge-base__kb_list_stories directly. If --tag or --plan-status were provided, pass them as plan_tag and plan_status:

mcp__knowledge-base__kb_list_stories({
  epic: "platform",
  states: ["ready_for_qa", "in_qa", "failed_qa", "failed_code_review", "in_review",
           "ready_for_review", "ready", "in_progress", "backlog",
           "completed", "cancelled", "deferred"],
  limit: N,
  ...(TAG ? { plan_tag: TAG } : {}),
  ...(PLAN_STATUS ? { plan_status: PLAN_STATUS } : {})
})

If the result contains 0 stories, go to Step 6 (Fallback).

Step 3 — Collect N Candidates via `mcpknowledge-basekb_get_next_story`

Call mcp__knowledge-base__kb_get_next_story up to N times directly (no sub-agent), accumulating excluded IDs. Pass plan_tag and plan_status if provided:

# First call — exclude all active worktrees
mcp__knowledge-base__kb_get_next_story({
  epic: "platform", include_backlog: true, exclude_story_ids: ACTIVE_IDS,
  ...(TAG ? { plan_tag: TAG } : {}),
  ...(PLAN_STATUS ? { plan_status: PLAN_STATUS } : {})
})
→ returns STORY_A

# Second call — exclude active + first result
mcp__knowledge-base__kb_get_next_story({
  epic: "platform", include_backlog: true, exclude_story_ids: [...ACTIVE_IDS, "STORY_A"],
  ...(TAG ? { plan_tag: TAG } : {}),
  ...(PLAN_STATUS ? { plan_status: PLAN_STATUS } : {})
})
→ returns STORY_B

# ... repeat until N items or no more candidates

Stop when mcp__knowledge-base__kb_get_next_story returns story: null or you have N items.

Step 4 — Re-sort by Priority Rank

After collecting candidates, re-sort by the rank table above (QA > Fix > Review > Dev > Elaboration > Creation), then by story priority within each rank:

priority order: critical (1) > high (2) > medium (3) > low (4) > null (5)

Important: kb_get_next_story only returns stories in ready or backlog state. Stories in other actionable states (ready_for_qa, failed_qa, failed_code_review, in_review, ready_for_review) must be pulled from the Step 2 kb_list_stories result and added to the candidates list before sorting.

To build the full candidate set:

From the kb_list_stories result, take all stories in inflight states: ready_for_qa, in_qa, failed_qa, failed_code_review, in_review, ready_for_review
From kb_get_next_story calls, collect up to N stories in ready or backlog state
Remove any ACTIVE_IDS from the combined list
Sort the combined list by rank, then by priority within rank
Take the top N

Step 5 — Map State to Command

For each story in the sorted list, determine the command based on its state field:

State	Category	Command Template
`ready_for_qa`	Needs QA	`/qa-verify-story {FEATURE_DIR} {STORY_ID}`
`in_qa`	Needs QA	`/qa-verify-story {FEATURE_DIR} {STORY_ID}`
`in_review`	Needs Fix	`/dev-fix-story {FEATURE_DIR} {STORY_ID}`
`failed_code_review`	Needs Fix	`/dev-fix-story {FEATURE_DIR} {STORY_ID}`
`failed_qa`	Needs Fix	`/dev-fix-story {FEATURE_DIR} {STORY_ID}`
`ready_for_review`	Needs Code Review	`/dev-code-review {FEATURE_DIR} {STORY_ID}`
`ready` or `in_progress`	Needs Dev	`/dev-implement-story {FEATURE_DIR} {STORY_ID}`
`backlog` (elaborated)	Needs Elaboration	`/elab-story {FEATURE_DIR} {STORY_ID}`
`backlog` (no story)	Needs Story Creation	`/pm-story generate {FEATURE_DIR} {STORY_ID}`

Distinguishing elaborated vs. not: A backlog story is "elaborated" if its story_dir exists on disk and contains a story.yaml file. If the directory or file doesn't exist, it needs story creation.

Feature directory derivation from story ID prefix:

Prefix	Feature Directory
`WINT`	`plans/future/platform/wint`
`KBAR`	`plans/future/platform/kb-artifact-migration`
`AUDT`	`plans/future/platform/code-audit`
`TELE`	`plans/future/platform/telemetry`
`INFR`	`plans/future/platform/infrastructure`
`LNGG`	`plans/future/platform/langgraph-update`
`MODL`	`plans/future/platform/model-experimentation`
`LERN`	`plans/future/platform/learning-loop`
`SDLC`	`plans/future/platform/sdlc-agents`
`AUTO`	`plans/future/platform/autonomous-dev`
`WKFL`	`plans/future/platform/workflow-learning`

If the story record has a story_dir field, use that to derive the feature directory (strip the story ID segment). If the prefix is not found in the table above, use plans/future/platform as the feature directory.

Step 6 — Fallback (DB Empty)

If Step 2 returned 0 stories, the database likely needs seeding.

Read plans/future/platform/WORK-ORDER-BY-BATCH.md directly
Parse the markdown tables to find unblocked stories
Output the results with a prominent warning:

⚠️  KB database has no stories — falling back to WORK-ORDER-BY-BATCH.md
    Run a DB seed to populate stories for reliable dependency resolution.

[results from markdown parsing]

When parsing the markdown fallback:

Stories with [x] checkbox or ✅ status are complete — skip them
Stories with ❌ or strikethrough are cancelled — skip them
Stories with 🔧 are being worked — skip them
For remaining stories, check if all (#NN) dependency references point to completed rows
Use the same state-to-command mapping, inferring state from status emoji:
- 🔍 → ready_for_qa (Needs QA)
- ⏸️ → in_review (Needs Fix — review findings exist)
- 📋 → ready_for_review (Needs Code Review)
- 🚧 → in_progress (Needs Dev)
- ⏳ → ready (Needs Dev)
- 📝 / 🆕 → backlog (Needs Elaboration or Story Creation)
- ⚪ → backlog (Needs Elaboration or Story Creation)
Apply the same stage-first sorting (QA > Fix > Review > Dev > Elaboration > Creation)

Step 7 — Output

Format the results as a table. If filters were applied, show them in the header. Inflight work appears first (per the rank order):

Next actions (N items):
Filters: tag=testing, plan-status=in-progress    ← only if filters were used

| # | Category          | Story                              | Priority | Command |
|---|-------------------|------------------------------------|----------|---------|
| 1 | Needs QA          | WKFL-002 — Story title             | high     | /qa-verify-story plans/future/platform/workflow-learning WKFL-002 |
| 2 | Needs Fix         | WINT-0030 — Story title            | high     | /dev-fix-story plans/future/platform/wint WINT-0030 |
| 3 | Needs Code Review | WKFL-007 — Story title             | medium   | /dev-code-review plans/future/platform/workflow-learning WKFL-007 |
| 4 | Needs Dev         | WKFL-010 — Story title             | medium   | /dev-implement-story plans/future/platform/workflow-learning WKFL-010 |
| 5 | Needs Elaboration | WKFL-008 — Story title             | low      | /elab-story plans/future/platform/workflow-learning WKFL-008 |

If stories were excluded due to active worktrees, note them below the table:

Excluded (active worktree): WKFL-006, WINT-0100

If no unblocked stories exist:

No actionable stories found.

All remaining stories are blocked by unresolved dependencies.
Use `kb_list_stories` to inspect blocked stories.

Edge Cases

Scenario	Behavior
DB returns 0 stories	Fall back to WORK-ORDER-BY-BATCH.md with warning
All stories blocked	Report "no actionable stories" with count of blocked
Story has no feature prefix match	Use `plans/future/platform` as feature dir
N > available unblocked stories	Return all available (less than N)
MCP tools unavailable	Fall back to WORK-ORDER-BY-BATCH.md with warning
`mcp__knowledge-base__worktree_list_active` fails	Treat ACTIVE_IDS as empty, proceed without exclusion
Story in `in_progress` with no active worktree	Treat as Needs Dev (rank 4)

next-actions