hashgraph-online/awesome-codex-plugins / plugins/beastlabai/multi-llm-plugin/skills/multi-llm
multi-llm
FlaggedOrchestrate plan reviews, task generation, implementation and code reviews across several LLM CLIs in parallel, or ask every model one free-text question about a plan. Use when the user says: review my plan, multi-llm, run a code review, generate tasks from a plan, implement this plan, ask the models about my plan.
schemas/cascade_record.schema.json
View on GitHub{
"$schema": "http://json-schema.org/draft-07/schema#",
"$id": "cascade_record.schema.json",
"title": "A --cascade's pointer and record ({plan}/<composite>/cascade-current.json, cascades/<stamp>-<id8>/record.json)",
"description": "utils/improve_cascade.py writes both files, atomically, and checks every read of them against this schema with utils/schema_validator.py (the draft-07 subset the plugin carries, so no runtime dependency). A file that fails is exit 3 naming the file, like an unreadable state.json. The pointer names the current cascade of one composite (improve-plan or improve-code); each cascade keeps its own directory, and starting a new one never renames or reuses a path: the new directory and its record are written first, an unfinished predecessor is marked `abandoned`, then the pointer is replaced. The two documents are told apart by their keys; the definitions `pointer` and `record` are what each read checks. The prose contract is docs/development/workflows.md, 'A cascade (`--cascade`)'.",
"oneOf": [
{"$ref": "#/definitions/pointer"},
{"$ref": "#/definitions/record"}
],
"definitions": {
"token": {
"type": "string",
"pattern": "^[0-9a-f]{16,64}$",
"description": "A random hex token: the cascade id, or a review round id."
},
"stamp": {
"type": ["string", "null"],
"description": "A local ISO timestamp, or null when the event has not happened."
},
"policy": {
"type": ["string", "null"],
"enum": ["expand", "preserve", null],
"description": "A --scope-policy value, or null for none (each round resolves its own)."
},
"pointer": {
"type": "object",
"additionalProperties": false,
"required": ["schema_version", "cascade_id", "dir"],
"properties": {
"schema_version": {"type": "integer", "const": 1},
"cascade_id": {"$ref": "#/definitions/token"},
"dir": {
"type": "string",
"pattern": "^cascades/[0-9]{8}T[0-9]{6}-[0-9a-f]{8}$",
"description": "The cascade's directory, relative to the composite's run dir: cascades/<YYYYmmddTHHMMSS>-<first 8 characters of cascade_id>."
}
}
},
"counts": {
"type": ["object", "null"],
"additionalProperties": true,
"description": "A closed stage's counts, as utils/cascade_archive.py computes them from the stage's published artifacts (stage_counts); null until the stage is closed. The archive owns the definitions; the types are pinned here.",
"properties": {
"findings": {"type": "integer"},
"valid": {"type": "integer"},
"invalid": {"type": "integer"},
"needs_decision": {"type": "integer"},
"reraises": {"type": "integer"},
"applied": {"type": "integer"},
"no_op": {"type": "integer"},
"declined": {"type": "integer"},
"skipped_by_user": {"type": "integer"},
"held_back": {"type": "integer", "description": "Scope deferrals of the stage's review round (absent in counts recorded before the column)."},
"judge_approved": {"type": "integer"},
"judge_salvaged": {"type": "integer"},
"judge_skipped": {"type": "integer"},
"nothing_to_apply": {"type": "boolean"},
"time_s": {"type": ["number", "null"]},
"cost": {"type": ["string", "null"], "description": "null is 'not recorded', never zero."}
}
},
"stage": {
"type": "object",
"additionalProperties": false,
"required": ["index", "model", "state", "bound_round", "scope_policy",
"started_at", "finished_at", "skip_reason", "counts"],
"properties": {
"index": {"type": "integer", "minimum": 1, "description": "1-based; equals the stage's position in `stages`."},
"model": {"type": "string", "minLength": 1, "description": "The stage's one canonical provider:model spec."},
"state": {
"type": "string",
"enum": ["pending", "running", "done", "skipped"],
"description": "pending until the stage's inputs are frozen and it is handed out; running from then; done once closed out (archived); skipped for a middle stage whose review produced no result."
},
"bound_round": {
"anyOf": [{"$ref": "#/definitions/token"}, {"type": "null"}],
"description": "The review round id most recently seen carrying this stage's provenance block; binding only asks whether it is set."
},
"scope_policy": {"$ref": "#/definitions/policy", "description": "The policy the stage started under."},
"started_at": {"$ref": "#/definitions/stamp"},
"finished_at": {"$ref": "#/definitions/stamp"},
"skip_reason": {"type": ["string", "null"], "description": "Why a skipped stage was skipped: its model's failure reason."},
"counts": {"$ref": "#/definitions/counts"}
}
},
"record": {
"type": "object",
"additionalProperties": false,
"required": ["schema_version", "cascade_id", "mode", "state", "started_at",
"scope_policy", "selectors", "stages", "finished_at"],
"properties": {
"schema_version": {"type": "integer", "const": 1},
"cascade_id": {"$ref": "#/definitions/token"},
"mode": {"type": "string", "enum": ["improve-plan", "improve-code"]},
"state": {"type": "string", "enum": ["running", "finished", "abandoned"]},
"started_at": {"type": "string"},
"scope_policy": {"$ref": "#/definitions/policy", "description": "The policy for stages not started yet: the typed --scope-policy at the start, or a later typed one."},
"selectors": {
"type": "object",
"additionalProperties": false,
"required": ["files_file", "base_ref"],
"description": "improve-code's typed selectors, frozen at the start so a bare continuation reproduces them. Both null on improve-plan.",
"properties": {
"files_file": {
"type": ["string", "null"],
"enum": ["files.txt", null],
"description": "files.txt in the cascade's directory (one repository-relative path per line) when --files or --files-from <file> was typed; every stage's review gets --files-from with its absolute path."
},
"base_ref": {"type": ["string", "null"], "description": "--base-ref as typed; every review and apply gets it."}
}
},
"stages": {
"type": "array",
"minItems": 1,
"items": {"$ref": "#/definitions/stage"}
},
"finished_at": {"$ref": "#/definitions/stamp", "description": "When the cascade finished or was abandoned."}
}
}
}
}
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| receiving-code-reviewobra/superpowersUse when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementation | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| requesting-code-reviewobra/superpowersUse when completing tasks, implementing major features, or before merging to verify work meets requirements | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| using-git-worktreesobra/superpowersUse when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallback | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| verification-before-completionobra/superpowersUse when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |