mcpbeat

Agents Consilium

codealive-ai/agents-consilium

Run external coding agents (Codex, Claude Code, OpenCode, native Grok Build, Gemini) as independent reviewers, repository researchers, or single-agent implementers. Use for multi-model opinions and code review, evidence-backed exploration of local or remote repositories, full-access delegation, steerable long-running work, or reattaching to delegated runs. Not for simple questions answerable directly from docs or the current codebase.

731k tokens
context cost
the whole folder, loaded on every use
76
files
ships runnable scripts
0
copies elsewhere
how many repositories repackaged it
115
stars on the repo
on the repository, not the skill itself

Install

one command, takes just this skill from the repository
npx skills add https://github.com/CodeAlive-AI/ai-driven-development --skill agents-consilium

What comes with it

2 383 209 bytes besides the instruction
ACP-RESEARCH.md
README.md
assets/agents-consilium-cover.png
config.example.json
config.json
prompts/broad-analyst.txt
prompts/broad-lateral.txt
prompts/explore.txt
prompts/judge.txt
prompts/probe-generic.txt
prompts/specialist.txt
references/configuration.md
references/delegate.md
references/explore.md
references/review.md
references/runtime-contracts.md
references/testing.md
scripts/consilium
scripts/lib/artifacts.sh
scripts/lib/backend_contract.py
scripts/lib/backend_run.sh
scripts/lib/code_review_validate.py
scripts/lib/common.sh
scripts/lib/config.sh
scripts/lib/debug_tape.py
scripts/lib/dedup-findings.py
scripts/lib/delegate.sh
scripts/lib/discovery-pass.sh
scripts/lib/events.py
scripts/lib/explore.sh
scripts/lib/human_id.py
scripts/lib/judge-runner.sh
scripts/lib/mode_policy.py
scripts/lib/normalize_stream.py
scripts/lib/progress.sh
scripts/lib/prompt_pipeline.py
scripts/lib/review_ask.sh
scripts/lib/review_code.sh
scripts/lib/review_super.sh
scripts/lib/review_ultra.sh

The instruction itself

4 sections, as written by the author

Agents Consilium

Use only scripts/consilium. Select the mode from the user's intent and load only the linked reference needed for that mode.

Decision map by intent

| User intent | Mode / reasonable default | Access | Read first |

|---|---|---|---|

| Independent architecture, design, debugging, or planning opinions | review ask with enabled profiles | read-only | references/review.md |

| Review a file or diff | review code (basic: security + correctness) | read-only | references/review.md |

| Review high-risk code | review code --depth specialists or super | read-only | references/review.md |

| Maximize review coverage | review code --depth ultra | read-only | references/review.md |

| Understand a local or remote repository | explore with Grok 4.5 | read-only | references/explore.md |

| Implement with one external worker | delegate -a <exact-id> (steerable by default) | full YOLO | references/delegate.md |

| Redirect or lifecycle-monitor a long-running worker | delegate; steer with --mode auto, observe with watch | full YOLO | references/delegate.md |

| Let work outlive the caller or reattach later | delegate --detach, then watch or wait | full YOLO for worker | references/delegate.md |

| Change profiles, effort, progress, limits, or artifacts | configuration | mode-dependent | references/configuration.md |

| Diagnose events, capabilities, policy, prompts, or workflows | runtime contract | mode-dependent | references/runtime-contracts.md |

| Run or extend tests | offline fake suite by default | test-dependent | references/testing.md |

review finds and validates problems. explore builds a relevance-first context map and answers from evidence. Do not substitute one for the other.

Default launch commands

scripts/consilium --list-agents

# Independent opinions
scripts/consilium review ask "Should we use Postgres or SQLite?"
scripts/consilium review ask -a codex,grok,claude-fable --prompt-file prompt.md

# Code review
scripts/consilium review code path/to/file.py
git diff HEAD | scripts/consilium review code --diff
scripts/consilium review code --depth specialists path/to/file.py
scripts/consilium review code --depth super path/to/file.py
scripts/consilium review code --depth ultra path/to/file.py

# Repository exploration
scripts/consilium explore "How is authentication wired up?"
scripts/consilium explore --repo owner/repository --ref main "How are plugins loaded?"

# Steerable delegate (default); run from the target project CWD
scripts/consilium delegate -a grok "Implement the caching layer and run tests."
scripts/consilium delegate steer run_<id> --mode auto "Keep the API compatible."
scripts/consilium delegate status run_<id> --json
scripts/consilium delegate watch run_<id>
scripts/consilium delegate wait run_<id>

# Explicit direct one-shot delegate
scripts/consilium delegate -a grok --one-shot "Implement a quick isolated task."

# Detached delegate and recovery
RUN_ID=$(scripts/consilium delegate -a grok --detach "Implement SPEC.md.")
scripts/consilium delegate list --active
scripts/consilium delegate watch "$RUN_ID"  # lifecycle only; no tool/file/text stream
scripts/consilium delegate wait "$RUN_ID"

steerable means the run accepts control commands; it does not imply rich

tool-level observability. watch is the supported lifecycle monitor. It shows

run, steer, and turn-boundary transitions plus heartbeats and terminal state,

but not the current tool, file, model text, or reasoning. Use wait to collect

the final answer. Do not inspect the private registry or audit.jsonl for

routine progress monitoring.

Detail map

| File | Load for |

|---|---|

| references/review.md | selection, effort overrides, depths, progress, output, exit codes |

| references/explore.md | sources, answer shape, isolation, trust boundary, web, provenance |

| references/delegate.md | YOLO rules, steering workflow, detach, mailbox states, delivery guarantees |

| references/configuration.md | prerequisites, profiles, shell-safe prompts, environment, limits |

| references/runtime-contracts.md | events, debug tape, safety/capabilities, workflows, prompt layers, artifacts |

| references/testing.md | offline suite and opt-in real-backend smoke tests |

| ACP-RESEARCH.md | deferred ACP transport research |

How to use it

Copy the folder

Take codealive-ai/agents-consilium from the repository into ~/.claude/skills for personal use, or into .claude/skills inside a project.

Check the name does not clash

The agent identifies a skill by the name field in its header. Two skills with the same name cannot sit side by side — one of them will be ignored.