Prompt Catalog
Prompt Catalog
This repository treats prompts as governed platform artifacts.
Prompt files live in prompts/ and should be selected by objective.
1) MultiAgent Review Prompt
File:
prompts/20260305-multiagent-review-prompt-jay.md
Use when:
- you need hostile repository review before execution
- you need strict ExecPlan-first enforcement
Expected outputs:
- JSON report with issues, missing plans, security findings, fixes
- concise human summary
2) Control Plane MVP Execution Prompt
File:
prompts/20260305-multiagent-control-plane-mvp-execution-jay.md
Use when:
- you are building the operational multiagent control plane
- you need referee + scoring + evidence pipeline behavior
Expected outputs:
- deterministic JSON status report
- concise summary (<= 12 lines)
3) Execute Merged Platform Backlog Prompt
File:
prompts/20260305-execute-merged-platform-backlog-jay.md
Use when:
- platform ExecPlans are merged but implementation is incomplete
- you need strict sequence execution and stop-on-blocker behavior
Required plan order:
20260304-platform-validator-jay-execplan20260304-todo-automation-jay-execplan20260304-repo-health-engine-jay-execplan20260304-spec-test-runner-jay-execplan
Expected outputs:
- deterministic JSON execution report
- concise summary (<= 12 lines)
4) Design Planner Control Plane Prompt
File:
prompts/20260310-design-planner-control-plane-codex-01.md
Use when:
- defining the planner control plane architecture
- defining session artifacts, graph schema, and contract projection rules
- defining players, moves, referee logic, and commitment stages for the planner game
- producing design-phase specs before implementation work begins
Expected outputs:
- architecture decisions with rationale
- explicit schema and contract proposals
- open questions with recommended disposition
5) Review Planner Control Plane Prompt
File:
prompts/20260310-review-planner-control-plane-codex-01.md
Use when:
- hostile-reviewing planner design artifacts
- checking for hidden state, weak authority boundaries, or vague command semantics
- verifying that provider sync remains downstream from canonical local state
- checking that game roles, illegal moves, and referee semantics remain coherent
Expected outputs:
- ordered findings with file references
- explicit assumptions and residual risks
Governance Notes
- Prompts must align with canonical plan key format.
- Canonical implementation logic is
src/platform_tools/*. - Human SSH-signed commits remain required for authoritative finalization.
Hostile Review Runbook
Use this runbook with prompts/20260305-multiagent-review-prompt-jay.md.
- Create a dedicated branch:
draft-execplan/hostile-review-<agent>-YYYYMMDD
- Execute hostile review only on that branch.
- Generate auditable artifacts in
artifacts/review/:hostile-review-report.jsonhostile-review-summary.mdvalidation-evidence.json
- Run deterministic checks in this order:
bin/execplan-validate .agent/execplans/*.mdbin/sync-todosbin/repo-health-checkbin/execplan-testbin/run-local-ci
- Commit artifacts and findings on the hostile-review branch and open a PR.