cursor-e2e

name: cursor-e2e description: Run end-to-end smoke tests for the Mycelium cursor adapter. Verifies cursor-agent prereqs, single-host dispatch, multi-host dispatch through the hub, cross-family negotiation with claude_code/openclaw, workspace asset drift, and auth failures. Use when validating the cursor integration on a fresh install, after touching cursor-family code (integrations/cursor/**, daemon/dispatch.py, daemon/runner.py), or after upgrading cursor-agent itself. argument-hint: "[--quick | --full | --multi-host]"

Cursor Adapter End-to-End Testing

Validate the cursor adapter by exercising the cold-spawn path end-to-end on real hosts. The general e2e skill covers stack health, memory, and negotiation; this one focuses on the cursor-specific surface that only cursor-agent exercises.

Arguments

• --quick — Prereqs + single-host dispatch only (< 1 min)
• --full — Quick + workspace asset drift + auth failure path (~ 3 min)
• --multi-host — Full + spoke dispatch through hub + cross-family negotiation (~ 8 min, requires SSH to spoke hosts)
• No argument — defaults to --full

Prerequisites

Before any phase runs, confirm:

# 1. cursor-agent on PATH
which cursor-agent
cursor-agent --version

# 2. Authenticated on this host. cursor-agent stores its access/refresh tokens
#    at ~/.config/cursor/auth.json (NOT ~/.cursor/cli-config.json — that file
#    holds session metadata only, the token field there has been removed).
ls ~/.config/cursor/auth.json
python3 -c "import json,os; p=os.path.expanduser('~/.config/cursor/auth.json'); j=json.load(open(p)); print('authenticated' if j.get('accessToken') else 'NOT LOGGED IN')"

# 3. Mycelium backend reachable + cursor doctor checks ok
mycelium doctor --mode auto

Fail criteria: any of these missing → run cursor-agent login and mycelium adapter add cursor --step=daemon before proceeding.

Phase 1: Single-host basic dispatch

The minimum viable cursor agent loop: create room → create cursor agent → @-mention → verify response posts back to the room.

# Setup
mkdir -p /tmp/cursor-e2e-workspace
mycelium room create cursor-e2e
mycelium daemon subscribe cursor-e2e

# Create a cursor agent
mycelium agent create cursor-x \
  --adapter cursor \
  --cwd /tmp/cursor-e2e-workspace \
  --room cursor-e2e \
  --description "smoke test agent"

# Verify workspace assets dropped
ls /tmp/cursor-e2e-workspace/.cursor/rules/mycelium.mdc
ls /tmp/cursor-e2e-workspace/AGENTS.md
grep -q "<!-- mycelium:start -->" /tmp/cursor-e2e-workspace/AGENTS.md && echo "marker present"

# Verify handle owned in daemon.toml
grep -A 5 '\[handles\]' ~/.mycelium/daemon.toml | grep cursor-x

# Invoke via the daemon
mycelium agent invoke cursor-x "Reply with just the string OK so I know you got this." --room cursor-e2e

# Wait for response (cursor cold-spawn is ~10-30s)
sleep 30
# Inspect recent chat messages (point-in-time read)
mycelium room messages cursor-e2e --limit 5

Fail criteria:

• No response in room within 60s → check ~/.mycelium/logs/daemon.log for dispatch @cursor-x lines
• "cursor-agent not authenticated" in logs → user needs to re-login
• Handle missing from daemon.toml → mycelium agent create didn't persist; check CursorIntegration.register
• Workspace assets missing → install_workspace_assets raised silently; check for NotADirectoryError

Phase 2: Workspace asset drift

Verify the adapter heals AGENTS.md when the user adds content outside the marker fence, and that drift is detected by mycelium doctor.

# Add user content OUTSIDE the mycelium marker block
cat > /tmp/cursor-e2e-workspace/AGENTS.md <<'EOF'
# My project agents

This is content I wrote myself. Mycelium should never touch it.

<!-- mycelium:start -->
(stale mycelium block placeholder)
<!-- mycelium:end -->

## More of my content
This is also mine.
EOF

# Force a re-register by removing + re-creating the agent
mycelium agent rm cursor-x --room cursor-e2e -y
mycelium agent create cursor-x --adapter cursor --cwd /tmp/cursor-e2e-workspace --room cursor-e2e

# Verify: my content preserved, mycelium block refreshed
grep -c "This is content I wrote myself" /tmp/cursor-e2e-workspace/AGENTS.md
grep -c "More of my content" /tmp/cursor-e2e-workspace/AGENTS.md
# The mycelium-managed block opens with `# Mycelium Agent` (asset header)
grep -c "# Mycelium Agent" /tmp/cursor-e2e-workspace/AGENTS.md

# Now nuke the .cursor dir and check that doctor surfaces the drift
rm -rf /tmp/cursor-e2e-workspace/.cursor
mycelium doctor --mode auto 2>&1 | grep -A 2 "cursor workspace assets"

Fail criteria:

• User content lost → _strip_agents_md_section is too aggressive; should only remove between markers
• Marker block not refreshed → _write_agents_md_section didn't run on re-register
• Doctor didn't flag the missing rule file → cursor doctor checks not wired up

Phase 3: Auth-failure friendly path

Simulate "user installed cursor-agent but never ran cursor-agent login" and verify the daemon posts an actionable error rather than a stack trace.

# Backup the auth file (this is where cursor-agent reads its tokens from at
# spawn time — not ~/.cursor/cli-config.json, that one only holds session
# metadata).
cp ~/.config/cursor/auth.json /tmp/auth.backup.json

# Move auth.json aside so cursor-agent fails to authenticate.
rm ~/.config/cursor/auth.json

# Try to invoke — should post a friendly message in the room, NOT crash the daemon.
mycelium agent invoke cursor-x "Anything" --room cursor-e2e
sleep 15
mycelium room messages cursor-e2e --limit 3

# Confirm doctor catches the bad state while it's still bad
mycelium doctor --mode auto 2>&1 | grep -A 2 "cursor-agent login"

# Restore auth — if the file got recreated, this overwrites it; otherwise mv is fine
cp /tmp/auth.backup.json ~/.config/cursor/auth.json
rm /tmp/auth.backup.json
# Verify
mycelium doctor --mode auto 2>&1 | grep "cursor-agent login"
# If for any reason the restore left auth in a weird state, run:
#   cursor-agent login

Fail criteria:

• Daemon crashed (check systemctl --user status mycelium-daemon) → auth detection raised instead of returning a SpawnResult
• Room shows a Python traceback → _detect_auth_required regression
• Doctor didn't flag the missing token → cursor login check not reading the right field

Phase 4: Multi-host dispatch (requires --multi-host)

Validate that a cursor agent created on a spoke responds to mentions sent from the hub.

Prerequisites: SSH access to spoke host (oclw3 / oclw5 / similar), mycelium client installed on spoke, spoke pointed at hub's backend.

HUB_HOST=oclw4  # hub
SPOKE_HOST=oclw3
ROOM=cursor-multi-e2e

# On the hub: create the room (room creation MUST happen on the host running the backend)
mycelium room create $ROOM

# On the spoke: install + create
ssh $SPOKE_HOST bash <<EOF
  # Verify spoke is pointed at the hub
  grep api_url ~/.mycelium/config.toml
  # Subscribe the spoke's daemon
  mycelium daemon subscribe $ROOM
  # Create the agent
  mkdir -p /tmp/cursor-spoke-ws
  mycelium agent create cursor-spoke \
    --adapter cursor --cwd /tmp/cursor-spoke-ws --room $ROOM \
    --description "spoke-side cursor agent"
EOF

# From the hub: subscribe locally + invoke the spoke agent
mycelium daemon subscribe $ROOM
sleep 3
# Cross-host invoke works because agent invoke falls through to the backend
# when the agent isn't in the local mirror (the hub didn't register cursor-spoke).
mycelium agent invoke cursor-spoke \
  "please reply with the hostname you're running on" \
  --room $ROOM

sleep 60  # cursor-agent cold start is slow
mycelium room messages $ROOM --limit 5
# Expect: a message FROM cursor-spoke containing the spoke host's hostname

Fail criteria:

• No reply from spoke → spoke daemon not subscribed or auth not propagated; check ssh $SPOKE_HOST mycelium daemon status
• Reply contains hub's hostname → handle ownership leaked to hub's daemon (both daemons firing on the same handle)
• Spoke daemon logs unknown adapter → spoke client out of date; uv tool install . --force --link-mode=copy from ~/mycelium/mycelium-cli

Phase 5: Cross-family negotiation (requires --multi-host)

Confirm a cursor agent and a counterparty (openclaw or claude_code) on different hosts can negotiate via IOC, autonomously — i.e. without an operator running mycelium negotiate respond between rounds. The daemon gained this autonomous-coordination path in 2026-05: the daemon polls /api/coordination-sessions, dynamically subscribes SSE to each active session sub-room, and on every coordination_tick cold-spawns the owned agent which then calls mycelium negotiate respond … itself.

ROOM=cursor-ioc-e2e

# Hub: create room, register the counterparty locally.
# (claude_code is the recommended counterparty when openclaw isn't running;
# the autonomous flow is identical for both adapters.)
mycelium room create $ROOM
mkdir -p /tmp/claude-hub-ws
mycelium agent create planner --adapter claude_code \
  --cwd /tmp/claude-hub-ws --room $ROOM \
  --description "ship-date-focused negotiation counterparty"

# Spoke: register cursor agent (auth must already be in place — cursor-agent
# stores tokens at ~/.config/cursor/auth.json on the spoke host).
ssh $SPOKE_HOST mycelium agent create designer --adapter cursor \
  --cwd /tmp/cursor-spoke-ws --room $ROOM \
  --description "design-polish-focused cursor side of the negotiation"

# Trigger negotiation (operator's only role: create + join × N).
mycelium session create -r $ROOM
mycelium session join --handle planner -m "Optimise for ship date" -r $ROOM
ssh $SPOKE_HOST mycelium session join --handle designer -m "Optimise for design polish" -r $ROOM

# Watch the autonomous flow play out. Within ~5s of each join the
# corresponding daemon's poller logs:
#   "dynamic subscribe → cursor-ioc-e2e:session:<short_id> (coordination session)"
# Within ~30s of join-window close, ticks arrive in the sub-room and the
# daemon dispatches:
#   "coordination_tick @planner — round=1 action=respond"
#   "dispatch @planner ← CognitiveEngine"
# Same on the spoke for @designer. Rounds advance every 10–30s as each
# agent posts a {"action":"accept"} / "reject" / {"offer":{...}}.

# Poll for terminal state. Typical converging session lands in 6–15 rounds
# (1–4 minutes); the cap is 20 rounds.
for i in $(seq 1 30); do
  STATE=$(curl -s "http://localhost:8000/api/coordination-sessions?limit=20" \
    | python3 -c "import sys,json
data=json.load(sys.stdin)
for s in data:
    if s['parent_room_name']=='$ROOM' and s['state'] in ('complete','failed'):
        print(s['state']); break
else: print('negotiating')")
  echo "[$i] state=$STATE"
  [ "$STATE" = "complete" ] || [ "$STATE" = "failed" ] && break
  sleep 15
done

# Inspect the consensus message
SHORT=$(curl -s "http://localhost:8000/api/coordination-sessions?limit=20" \
  | python3 -c "import sys,json
for s in json.load(sys.stdin):
    if s['parent_room_name']=='$ROOM' and s['state']=='complete':
        print(s['short_id']); break")
mycelium --json room messages "$ROOM:session:$SHORT" --type coordination_consensus --limit 200 \
  | python3 -c "import sys,json
for m in json.load(sys.stdin)['messages']:
    print(m['content']); break"
# Expect: {\"plan\":\"...\",\"assignments\":{...},\"broken\":false}

Fail criteria:

• No dynamic subscribe log on either daemon within 10s of join → daemon didn't reach the poller branch (deployment regression — verify the installed dispatch.py contains poll_coordination_sessions)
• dynamic subscribe fires but no dispatch @<handle> on tick → handle not in this daemon's daemon.toml.handles (sibling daemon owns it on a different host); confirm with mycelium daemon ls
• dispatch fires but spawn exits 1 with Credit balance is too low (claude) → counterparty's API key is unfunded; switch counterparty or top up
• dispatch fires but spawn exits 1 with SessionEnd hook ... not found (claude) → stale ~/.claude/settings.json from an older mycelium-cli; rerun mycelium adapter add claude-code --reinstall
• Cursor agent never produces a counter_offer / accept → cursor-agent isn't seeing the mycelium rules (workspace assets missing on spoke); rerun mycelium agent create designer … to redrop them
• broken: true in consensus after 20 rounds → opposing personas locked on incompatible positions; an agent-prompt issue, not coordination — re-run with less polarised intent strings or a higher n_steps budget on the CFN side
• counter_offer_not_your_turn loops → agent ignored the per-round allowed_actions in the tick payload; agent-prompt issue

Cleanup

# Hub
mycelium agent rm cursor-x --room cursor-e2e --full -y 2>/dev/null
mycelium agent rm planner --room cursor-ioc-e2e --full -y 2>/dev/null
mycelium daemon unsubscribe cursor-e2e 2>/dev/null
mycelium daemon unsubscribe cursor-multi-e2e 2>/dev/null
mycelium daemon unsubscribe cursor-ioc-e2e 2>/dev/null
rm -rf /tmp/cursor-e2e-workspace

# Spoke (if --multi-host ran)
ssh $SPOKE_HOST mycelium agent rm cursor-spoke --room cursor-multi-e2e --full -y 2>/dev/null
ssh $SPOKE_HOST mycelium agent rm designer --room cursor-ioc-e2e --full -y 2>/dev/null
ssh $SPOKE_HOST rm -rf /tmp/cursor-spoke-ws

# Drop rooms
curl -s -X DELETE http://localhost:8000/api/rooms/cursor-e2e
curl -s -X DELETE http://localhost:8000/api/rooms/cursor-multi-e2e
curl -s -X DELETE http://localhost:8000/api/rooms/cursor-ioc-e2e

Interpreting Failures

When to Update This Skill

• New cursor-specific feature in integrations/cursor/ (e.g. new cursor-agent flag, new asset dropped) → add a phase that exercises it
• New daemon-side regression caught in production → add a fail-criteria row
• New adapter family that can negotiate with cursor → add a cross-family phase