Epic Manager

Monitor an epic's dependency graph against live GitHub + K8s state and prepare the next wave of stories for dispatch.

Source of truth: The [Epic] tracker issue on GitHub. No local YAML files. The tracker issue body contains the dependency graph table and tracked issues checklist.

Input

/epic-manager <tracker-issue-number> [--dispatch]
/epic-manager <tracker-issue-number> --add <N> --blocked-by <N>[,N...] [--wave <N>] [--note "..."]

Help

If no issue number is provided, or --help / -h is passed, display this help text directly (do NOT run the script):

/epic-manager — Epic dependency graph tracker + dispatch advisor

USAGE
  /epic-manager <N>                          Show dashboard for epic tracker #N
  /epic-manager <N> --dispatch               Dashboard + copy-paste dispatch commands
  /epic-manager <N> --json                   Machine-readable JSON output
  /epic-manager <N> --add <M> --blocked-by <X>[,Y]   Add story #M to epic graph

DASHBOARD EXAMPLES
  /epic-manager 1526                         Live status of Odin's Spear epic
  /epic-manager 1527 --dispatch              Dashboard + ready dispatch commands

ADD STORY EXAMPLES
  /epic-manager 1526 --add 1507 --blocked-by 1495
  /epic-manager 1526 --add 1508 --blocked-by 1495 --wave 3

DASHBOARD ICONS
  ✅  done      — GitHub issue is CLOSED
  🔄  running   — Active K8s job found for this issue
  🟢  ready     — All blockers closed, not yet running
  🔴  blocked   — One or more blockers still OPEN

SOURCE OF TRUTH
  The [Epic] tracker issue body on GitHub contains the dependency graph.
  No local YAML files — GitHub is the single source of truth.

Speed Rules (UNBREAKABLE)

1. One tool call. Run the MJS script directly — no pre-fetching, no narration between calls.
2. Never re-fetch what the script already fetches. The script queries GitHub and K8s itself.
3. If the issue is not an [Epic] tracker, tell Odin to check the issue number.
4. If no issue number or --help, display the Help section above — do NOT run the script.

Execution

node .claude/skills/epic-manager/epic-manager.mjs <N> [--dispatch]

That is the entire execution. No other tool calls needed unless the output requires follow-up action (e.g., dispatching a ready story).

What the Script Does

Phase 1 — Load epic graph from GitHub

Fetches the [Epic] tracker issue body via gh issue view <N>. Parses the dependency graph markdown table:

| Wave | Issue | Title | Depends On |
|------|-------|-------|------------|
| 0 | #1516 | Migrate trial store | — |
| 1 | #1519 | Remove Redis client | #1516, #1517, #1518 |

Phase 2 — GitHub state sync

For every issue number in the graph (stories + their blockers), calls:

gh issue view <N> --json number,state,title

Phase 3 — K8s job sync

Queries active GKE jobs in fenrir-agents namespace:

kubectl get jobs -n fenrir-agents -o json

Only jobs with status.active > 0 are counted as running.

Phase 4 — Status computation

Phase 5 — Dashboard output

Wave-by-wave table with icons (same format as before).

--add mode

When --add is used, the script updates the tracker issue body directly on GitHub:

• Adds a new row to the dependency graph table
• Adds the issue to the tracked issues checklist
• No local files modified

After the Dashboard

Always use AskUserQuestion when asking Odin anything. Never output a question as plain text and wait — use the AskUserQuestion tool so Odin gets a proper prompt.

Always show a summary before asking. Before every AskUserQuestion, output a brief markdown summary of the issues in question so Odin has context without needing to look them up. Format:

**Ready to dispatch (Wave N):**
- #X — <title> (type/priority) — <1-line summary from issue body>
- #Y — <title> (type/priority) — <1-line summary from issue body>

or for stalls:

**Epic stalled — nothing ready or running:**
- N done, N blocked, 0 ready, 0 running
- Next unblock: #X needs #Y to close first

Then immediately follow with the AskUserQuestion call. The summary and the question should appear in the same response — no extra round-trip.

Read the output and:

1. If stories are ready — summarize the ready issues with titles, types, and priorities, then use AskUserQuestion: "Wave N is unblocked. Dispatch?" Options: "Dispatch #X", "Dispatch all in parallel", "Skip". Then invoke /dispatch or /fire-next-up #X #Y accordingly.
2. If nothing is ready and nothing is running — summarize the blockage chain, then use AskUserQuestion to flag the stall. Options: "Re-check", "Close epic".
3. If epic is complete — congratulate and close the tracker issue immediately (do NOT ask for confirmation):
   bashCopy

   gh issue close <N> --comment "🎉 Epic complete — all stories done."
   

Rules

1. Always run the script — never manually reconstruct the dashboard from memory.
2. K8s query is best-effort — if kubectl is unavailable, the script continues without K8s data.
3. Never dispatch without showing the dashboard first — Odin must see the state before agents are spawned.
4. GitHub is the source of truth — the tracker issue body contains the dependency graph. No local YAML files.
5. --add updates GitHub directly — the tracker issue body is edited in-place via gh issue edit.