DR API Execution Bootstrap
Use this as an installer and execution-enforcer skill.
When the user asks to apply this skill, apply it immediately. Do not ask whether to enforce the policy.
Apply to this workspace
- Inspect workspace startup/default files.
- Persist the execution policy in workspace bootstrap files (
AGENTS.md, MEMORY.md, or equivalent). - Patch surgically and preserve unrelated instructions.
- Validate with the strongest safe real test available.
- Report either
Configured and validated or Configured, but blocked by: .
For the concrete application checklist, read references/APPLY.md.
Enforcement contract
1) Execution policy
Set and enforce these defaults for future sessions:
- prefer direct in-session API execution
- do not spawn subagents unless the user explicitly asks
- default to fast mode single-run chain
- do one upfront preflight only:
- auth/token availability
- app code / function key / required secret availability
- one sanity endpoint check
- after preflight passes, execute the full API chain continuously without unnecessary pauses
2) Communication policy
Set and enforce:
- keep responses concise
- do not narrate every API call unless the user asks
- for blocked execution, report the blocker briefly and precisely
- for write operations, show one concise batch preview and wait for approval before executing
3) Operational execution rules
Read references/EXECUTION-PLAYBOOK.md and follow it for:
- bulk reads
- bulk writes
- failure handling
- resume behavior
- verification strategy
4) API-specific guidance
If the workflow involves Ikara-style CRUD/integration/service-compliance APIs, also read references/IKARA-PATTERNS.md before executing.
5) Validation requirements
After applying the rules, immediately validate them.
If safe and permitted, run one small real dev test and confirm:
- direct API call path works
- no subagent was spawned
- preflight + full-chain behavior is active
If real execution is not possible, run the strongest safe validation available and state exactly what prevented full validation.
6) Limits
If permissions, secrets, or tool access are missing:
- do not pretend they were enabled
- do not claim success
- report exactly what is missing
- keep the enforced policy in startup files anyway, unless file-write access is blocked
Reuse on other agents
If the user wants to replicate this behavior elsewhere, read references/INSTALL.md and provide the installation command plus the recommended bootstrap prompt.