| name | standup |
| description | Status snapshot for active agent team work — reads TaskList and each agent's
context.md, produces a table of who is doing what, what's blocked, and what's
ready to claim. Use during Phase 3 implementation to monitor parallel agent
progress. Usage: /standup [TICKET-ID]
|
Standup
Get an instant status snapshot of active agent team work. Reads the task system and each
agent's working memory to surface blockers, stale work, and next actions — without
interrupting agents mid-task.
Arguments
/standup — show all active tasks across all tickets/standup <TICKET-ID> — filter to tasks for a specific ticket (e.g. POC-123)
Step 0: Resolve project root
_gc=$(git rev-parse --git-common-dir 2>/dev/null) \
&& MROOT=$(cd "$(dirname "$_gc")" && pwd) \
|| MROOT=$(pwd)
WTROOT=$(git rev-parse --show-toplevel 2>/dev/null || pwd)
PDH=$( [ -f skills/plugin-dir.sh ] && pwd || find ~/.claude/plugins/cache -path '*/dev-team/*/skills/plugin-dir.sh' 2>/dev/null | sort -V | tail -1 | xargs -r dirname | xargs -r dirname )
dag-lib.sh ships in the plugin, not the user's repo, so it is resolved through
bash "$PDH/skills/plugin-dir.sh" file skills/orchestrate/dag-lib.sh. Each SKILL
bash block runs as a fresh shell, so Step 4 re-emits this stanza.
Step 1: Read task state
Read both views of task state and reconcile them:
- Call
TaskList for the in-session view.
- Read
.claude/tasks/*.json for the file-store view (the source of
truth that survives compaction):
_gc=$(git rev-parse --git-common-dir 2>/dev/null)
&& MROOT=$(cd "$(dirname "$_gc")" && pwd)
|| MROOT=$(pwd)
while IFS= read -r f; do
[ -f "$f" ] || continue
jq -r '[.task_id, .status, .subject // ""] | @tsv' "$f"
done < <(find "$MROOT/.claude/tasks" -maxdepth 1 -name '*.json' -type f 2>/dev/null)
If a TICKET-ID was provided, filter to tasks whose subject contains that ID.
If no tasks are found in either view: print `No active tasks found.` and stop.
Group tasks by status:
- `pending` — not yet claimed
- `in_progress` — claimed, actively being worked
- `completed` — done
- `blocked` — explicitly marked blocked
When TaskList and the file store disagree on a task's status, prefer
the file store (TaskList can stale out across `/clear` and across
async-spawned agents whose completion never closed the parent's
TaskList row).
---
## Step 2: Read agent context files
For each task with status `in_progress`, read the owning agent's context file:
```bash
_gc=$(git rev-parse --git-common-dir 2>/dev/null) \
&& MROOT=$(cd "$(dirname "$_gc")" && pwd) \
|| MROOT=$(pwd)
WTROOT=$(git rev-parse --show-toplevel 2>/dev/null || pwd)
cat $WTROOT/.claude/memory/<owner>/context.md 2>/dev/null
If no context file exists for an agent, note: (no context.md — agent hasn't written progress yet).
Also check for recent commits by the agent:
git log --oneline --since="2 hours ago" --author="<agent-name>" 2>/dev/null | head -5
(Agent name may appear in commit Co-Authored-By lines — grep for it if needed.)
Step 3: Assess staleness
For each in_progress task, determine if it looks stale:
Stale indicators (any one → flag as STALE):
- context.md was not updated in the last 30 minutes (check file mtime if possible)
- No commits from this agent in the last hour
- context.md says "blocked" or "waiting" without a corresponding SendMessage to Tech Lead
- Task has been
in_progress since before the last completed task finished
Flag stale tasks with ⚠️ STALE in the output.
Probably-completed-but-unmarked detection. Async-spawned agents
finish in their own sandbox session, so their TaskUpdate(completed)
never reaches the orchestrator. Their TaskList row stays
in_progress indefinitely while the work is actually done. Flag a
task as 🟡 LIKELY-DONE (separate from STALE) when ALL of:
- Status is
in_progress AND
- Owner has no live agent process (no recent context.md write, no
recent commits) AND
- The agent's expected output marker exists — either
.claude/tasks/<id>.json shows status: completed in the file
store but TaskList disagrees, OR the agent has commits referencing
the ticket within the last 2 hours but no activity since.
Surface these in a dedicated section — they need an orchestrator
TaskUpdate(<id>, completed) to close the loop and unblock dependent
tasks.
Step 4: Identify ready-to-claim tasks
For each pending task, compute readiness from the task store:
PDH=$( [ -f skills/plugin-dir.sh ] && pwd || find ~/.claude/plugins/cache -path '*/dev-team/*/skills/plugin-dir.sh' 2>/dev/null | sort -V | tail -1 | xargs -r dirname | xargs -r dirname )
DAG_LIB=$(bash "$PDH/skills/plugin-dir.sh" file skills/orchestrate/dag-lib.sh)
READY_TASKS=$(bash "$DAG_LIB" ready-set)
- A task is READY when its
task_id appears in the output of dag-lib.sh ready-set.
- A task is WAITING when it is
pending but NOT in the ready-set.
For each WAITING task, show the dependency chain by reading its task file:
_gc=$(git rev-parse --git-common-dir 2>/dev/null) \
&& MROOT=$(cd "$(dirname "$_gc")" && pwd) \
|| MROOT=$(pwd)
TASK_FILE="$MROOT/.claude/tasks/<task_id>.json"
DEP_IDS=$(jq -r '.depends_on // [] | .[]' "$TASK_FILE" 2>/dev/null)
For each dep_id in depends_on (reuse $DAG_LIB, or re-resolve if running
this block fresh):
PDH=$( [ -f skills/plugin-dir.sh ] && pwd || find ~/.claude/plugins/cache -path '*/dev-team/*/skills/plugin-dir.sh' 2>/dev/null | sort -V | tail -1 | xargs -r dirname | xargs -r dirname )
DAG_LIB=$(bash "$PDH/skills/plugin-dir.sh" file skills/task-dag.sh)
STATUS=$(bash "$DAG_LIB" status-of "$dep_id")
- Show:
WAITING on: <dep_id> (<STATUS>)
- If the dep's task file is missing: show
WAITING on: <dep_id> (file missing)
Use jq '.depends_on // []' as the default to ensure backward compatibility with task files that predate the depends_on field.
Step 5: Print standup report
Each task row includes an optional routing indicator showing where the task ran:
| Indicator | Meaning |
|---|
[local] | Task ran on the local agent — orchestrator companion record has saved_est_tokens > 0. |
[local→escalated] | Task started local but Claude had to take over — companion record has saved_est_tokens == 0. |
| (blank) | No companion record for this ticket (Claude ran it normally, or the wrapper fell back). |
Signal source: .claude/local-agent/metrics.jsonl contains two record types:
run.sh records — one per terminal path; fields: ts, outcome, exit_code,
saved_est_tokens, spent_tokens. No ticket field — these records are not keyed
by ticket and are ignored by the routing indicator.- Orchestrator companion records (written by
emit-orch-metric.sh) — fields include
ticket and saved_est_tokens (integer). This is the sole signal source for the
routing indicator: saved_est_tokens == 0 means the local agent ran but escalated to
Claude; any positive value means the local agent completed the task.
If the file is absent, no companion record exists for the ticket, or jq is unavailable,
the column is blank — best-effort and gracefully absent when the local-agent feature
hasn't run.
_gc=$(git rev-parse --git-common-dir 2>/dev/null) \
&& MROOT=$(cd "$(dirname "$_gc")" && pwd) \
|| MROOT=$(pwd)
METRICS="$MROOT/.claude/local-agent/metrics.jsonl"
routing_label() {
local ticket="$1"
[ -f "$METRICS" ] || { echo ""; return; }
command -v jq >/dev/null 2>&1 || { echo ""; return; }
local saved
saved=$(jq -r --arg t "$ticket" \
'select(.ticket == $t) | .saved_est_tokens' \
"$METRICS" 2>/dev/null | tail -1)
[ -z "$saved" ] || [ "$saved" = "null" ] && { echo ""; return; }
if [ "$saved" = "0" ]; then
echo "[local→escalated]"
else
echo "[local]"
fi
}
Standup — <TICKET-ID or "all tickets"> — <current time>
─── In Progress ──────────────────────────────────────────────────────
id:<N> [ic4] [local] Task 1 — cache layer (GetAllForFolder)
Last: "Wrote 3 tests, implementing GetAllForFolder now"
Commits: feat: POC-123 — add CachedAnalysis struct (14 min ago)
Status: ✅ on track
id:<N> [ic5] [local→escalated] Task 2 — export package (Exporter+impls)
Last: "Defined Exporter interface, sent to @ic4 and @qa. Writing CSV next."
Commits: none yet
Status: ⚠️ STALE — no commits in 45 min, check context
─── Pending (ready to claim) ─────────────────────────────────────────
id:<N> [unassigned] Task 4 — QA acceptance tests
Note: "Start after IC5 defines interface" — interface defined ✅ READY
─── Pending (blocked) ────────────────────────────────────────────────
id:<N> [unassigned] Task 3 — UI wiring
Waiting on: Task 1 (in_progress), Task 2 (in_progress)
─── Completed ────────────────────────────────────────────────────────
id:<N> [ic4] [local] Task 0 — spec review ✓
─── Summary ──────────────────────────────────────────────────────────
In progress: 2 | Ready to claim: 1 | Blocked: 1 | Done: 1
─── Suggested actions ────────────────────────────────────────────────
⚠️ ic5 Task 2 looks stale — check .claude/memory/ic5/context.md or SendMessage to @ic5
🟡 Task 5 likely-done — orchestrator should TaskUpdate id:5 → completed
✅ Task 4 is ready — @qa can claim via TaskUpdate now
⏳ Task 3 unblocks when Tasks 1+2 complete — Tech Lead should monitor
Omit any section that has no tasks. Omit the routing indicator from a row when the
signal is unavailable — do not show a blank [] placeholder.
Step 5.5: Epic rollup (SPEC-025 / M10)
When any epic under $MROOT/.claude/epics/ has non-completed children, surface
an epic section sourced from state.json via epic-lib.sh (not prose). Omit
the entire section when rollup is empty.
_gc=$(git rev-parse --git-common-dir 2>/dev/null) \
&& MROOT=$(cd "$(dirname "$_gc")" && pwd) \
|| MROOT=$(pwd)
PDH=$( [ -f skills/plugin-dir.sh ] && pwd || find ~/.claude/plugins/cache -path '*/dev-team/*/skills/plugin-dir.sh' 2>/dev/null | sort -V | tail -1 | xargs -r dirname | xargs -r dirname )
EPIC_LIB="$PDH/skills/epic/epic-lib.sh"
ROLLUP=$(bash "$EPIC_LIB" rollup)
If $ROLLUP is non-empty, print:
─── Epics ────────────────────────────────────────────────────────────
<epic_id> <title> mode:<execution_mode>
pending:N in_progress:N completed:N blocked:N / total:N
ready: <id>, <id>, …
waves: Wave 1: … → Wave 2: …
Each stdout line from rollup is one JSON object — pretty-print with jq when
available. Counts and ready set come only from that JSON (state.json), never
from free-text memory.
Step 6: Escalation check
After the report, check if any escalation is warranted:
Auto-escalate to Tech Lead if:
- Any task is STALE with no recent SendMessage to Tech Lead
- Two or more tasks are blocked waiting on the same dependency
- A completed task's output hasn't been consumed by the downstream task after 30+ min
If escalation is needed, add at the bottom:
─── Escalation needed ────────────────────────────────────────────────
Recommend: SendMessage to @tech-lead — ic5 Task 2 stale, may need unblocking
Do NOT send the message automatically — surface it for the engineer to decide.
Error Handling
- No tasks at all:
No tasks found. Run /kickoff <TICKET-ID> to create a task graph.
- TaskList unavailable (orchestration not initialized):
Agent Teams not initialized. Run /init-orchestration first.
- context.md missing for in_progress agent: note it but don't fail — the agent may not have written it yet
- Git log unavailable (not in a git repo): skip commit staleness check, rely on context.md mtime only
