Agentbase Teardown

AgentBase Teardown

Guided cleanup of all AgentBase resources for a project or agent.

Authentication

Read the shared auth setup reference at /agentbase skill's references/auth-setup.md for full IAM credential configuration. In brief: run bash .claude/skills/agentbase/scripts/check_credentials.sh iam to verify credentials are configured. NEVER read .greennode.json or .env directly — always use the helper scripts. If check_credentials.sh iam returns MISSING, STOP — you MUST read the "If Credentials Are Not Found" section in /agentbase skill's references/auth-setup.md and follow it exactly. Do NOT skip this or provide your own credential setup instructions.

Interaction Guidelines

• Never assume API response structure — always inspect the actual response first before extracting or filtering data. Do not guess field names.
• ALWAYS show full deletion plan before executing — never delete anything without showing the user exactly what will be removed
• ALWAYS require explicit confirmation (HARD GATE) — the user must respond with an explicit confirmation keyword (yes, confirm, ok, approve, proceed, go ahead, do it, lgtm, or equivalent affirmative) before any deletion begins. If the user responds with ANYTHING ELSE (deselecting items, questions, adjustments, or ambiguous text), treat it as additional input — update the plan and re-present for confirmation again. NEVER interpret a non-confirmation response as approval
• Support --dry-run — if the user passes --dry-run, show the plan only and do not execute any deletions
• Let user deselect items — after presenting the plan, let the user choose which items to keep (e.g., "keep the AIP key, delete everything else")
• Warn about shared resources — if a resource may be used by other agents (e.g., an AIP API key, an auth provider), explicitly warn the user (e.g., "This AIP key may be used by other agents")
• Show progress during deletion — report each deletion as it completes
• Always read full API response body — when calling platform APIs, capture and read the full JSON response (not just status codes). This avoids misidentifying field names or data structures, ensures correct field extraction, and enables better error handling and debugging.

How It Works

Step 1: Identify the Project

Determine which project/agent to tear down:

• If a project name is provided as argument, use it
• If .agentbase-state.json exists in the current directory, read the agent/project name from it
• Otherwise, ask the user which agent/project to tear down

Step 2: Discover Related Resources

Discover all resources matching the project name:

bash .claude/skills/agentbase/scripts/discovery.sh json

This returns JSON with all resources across all services. Parse the output to find resources matching the project name.

Resource matching priority:

1. Exact resource IDs from .agentbase-state.json or .greennode.json (preferred — most precise)
2. Prefix matching — look for resources whose name starts with the project name (case-insensitive). For example, project "my-agent" matches "my-agent-identity" and "my-agent-key", but NOT "test-my-agent".
3. Exact name match — if prefix matching returns no results, try exact match on the project name itself.

Warning: Always show exact resource IDs and full names in the deletion plan so the user can verify which resources will be affected. If matching finds resources that may belong to other projects, explicitly warn the user and ask them to confirm each resource individually.

Step 3: Present Deletion Plan

Show the user a numbered plan of what will be deleted:

Teardown Plan for "my-agent":
  1. Delete runtime endpoints (2 endpoints)
  2. Delete runtime "my-agent-rt" (v3)
  3. Delete auth provider "openai-key" (API Key)
  4. Delete agent identity "my-agent"
  5. Delete memory "my-agent-memory"
  6. Delete CR images for this project (3 images, all artifacts)
  7. Delete AIP API key "my-agent-key" (shared resource)

All deletions are IRREVERSIBLE.
Proceed with all? Or type numbers to exclude (e.g., "skip 6,7"):

If no related resources are found, tell the user and stop.

Step 4: Get User Confirmation

Wait for the user to:

• Confirm all deletions
• Deselect specific items (e.g., "skip 7" or "keep the AIP key")
• Cancel entirely

Do NOT proceed without explicit confirmation.

Step 5: Execute Deletions in Dependency Order

Delete in this specific order to avoid dependency errors. If any script call returns an auth error (401/403) during the teardown sequence, re-authenticate with bash .claude/skills/agentbase/scripts/get_token.sh --force and retry the failed call once. If the retry also fails, report the error and continue with the next deletion.

Phase 1 — Runtime endpoints (MUST run before Phase 2):

Why order matters: The API rejects runtime deletion if custom endpoints still exist. You must delete all non-DEFAULT endpoints before deleting the runtime.

# List endpoints for each runtime
bash .claude/skills/agentbase/scripts/runtime.sh endpoints list $RUNTIME_ID

# Delete each non-DEFAULT endpoint
bash .claude/skills/agentbase/scripts/runtime.sh endpoints delete $RUNTIME_ID $ENDPOINT_ID

Phase 2 — Runtimes (only after all custom endpoints are deleted):

bash .claude/skills/agentbase/scripts/runtime.sh delete $RUNTIME_ID

Phase 3 — Auth providers (API Key, Delegated, OAuth2):

# API Key providers
bash .claude/skills/agentbase/scripts/auth.sh apikey delete --name $NAME

# Delegated providers
bash .claude/skills/agentbase/scripts/auth.sh delegated delete --name $NAME

# OAuth2 providers
bash .claude/skills/agentbase/scripts/auth.sh oauth2 delete --name $NAME

Phase 4 — Agent identity:

bash .claude/skills/agentbase/scripts/identity.sh delete $NAME

Phase 5 — Memory:

bash .claude/skills/agentbase/scripts/memory.sh delete $MEMORY_ID

Phase 6 — Container Registry images:

Important: Each user has one pre-provisioned repository that cannot be deleted. Teardown only removes images (and their artifacts) belonging to the project. List images first, filter by project name, then delete each one.

# Step 1: List images, filter by project name substring (paginate if totalPage > 1)
PAGE=1
while true; do
  RESULT=$(bash .claude/skills/agentbase/scripts/cr.sh images list --name $PROJECT_NAME --page $PAGE --size 100)
  # For each image in $RESULT.data[].name:
  #   bash .claude/skills/agentbase/scripts/cr.sh images delete --name $IMAGE_NAME
  # (image delete cascades to all artifacts/tags under that image.)
  TOTAL_PAGE=$(echo "$RESULT" | jq '.totalPage // 1')
  [ "$PAGE" -ge "$TOTAL_PAGE" ] && break
  PAGE=$((PAGE + 1))
done

# Step 2: Verify project images are gone
bash .claude/skills/agentbase/scripts/cr.sh images list --name $PROJECT_NAME

Phase 7 — AIP API keys (optional, may be shared):

bash .claude/skills/agentbase/scripts/aip.sh api-keys delete $KEY_NAME

The aip.sh api-keys delete command sends the DELETE request and returns immediately. Poll with aip.sh api-keys get $KEY_NAME until you get a 404 to confirm deletion.

Step 6: Report Results

Show a summary of what was deleted and any errors:

Teardown Results for "my-agent":
  Deleted runtime endpoints (2)
  Deleted runtime "my-agent-rt"
  Deleted auth provider "openai-key"
  Deleted agent identity "my-agent"
  Deleted memory "my-agent-memory"
  Skipped CR images for "my-agent" (user chose to keep)
  Failed to delete AIP key "my-agent-key" (403 Forbidden)

Teardown finished. 5 of 7 resources removed. 2 failed — see errors above.

Step 7: Clean Up Local State

If .agentbase-state.json exists in the current directory, reset the wizard_step to 0 and clear resource IDs that were deleted (e.g., runtime_id, memory_id, agent_identity, aip_key_name, cr_repo_name). This prevents /agentbase-wizard resume from trying to resume with stale references to deleted resources. Only clear fields for resources that were actually deleted — keep fields for resources the user chose to skip/keep.

If .agentbase/ directory exists, offer to remove it:

Found local AgentBase files:
  - .agentbase-state.json (wizard state — will be reset)
  - .agentbase/ (token cache, temp files)
Reset wizard state and remove cache? (y/n)

If file operations fail (e.g., permission denied), report the specific error and suggest the user handle them manually.

Troubleshooting

Instructions

1. Parse the user's argument for project name and --dry-run flag.
2. Authenticate (see Authentication section).
3. Identify the project (argument, .agentbase-state.json, or ask user).
4. Discover all related resources using bash .claude/skills/agentbase/scripts/discovery.sh json.
5. Present the deletion plan with numbered items.
6. If --dry-run, stop after showing the plan.
7. Wait for user confirmation (allow deselecting items).
8. Execute deletions in dependency order using the script commands, reporting progress.
9. Show final summary.
10. Offer to clean up .agentbase-state.json and .agentbase/ directory if present.