Hiveworkflow

/hiveworkflow

Run a governed hive workflow for:

$ARGUMENTS

/hiveworkflow is the end-to-end mode for tasks where coordinated specialized agents materially improve the outcome. It is not the default path; plain /workflow remains better for tightly coupled work.

For file-changing hive work, /opusworkflow is the default outer route and /hiveworkflow is the inner contract. Direct /hiveworkflow invocation remains valid, but it must inherit the Claude/Opus planner-reviewer plus MiniMax-M2.7-highspeed executor policy before mutation.

outer_route: opusworkflow
inner_contract: hiveworkflow

Core Contract

For file-changing work, follow this order:

taste gate -> metacognitive route -> hive eligibility -> deepresearch
-> code audit -> role map -> blackboard -> pre-plan /introspect -> SPEC.md
-> packet execution via /parallel when eligible -> aggregation
-> post-synthesis /introspect -> /verify -> closeout

Do not edit files before SPEC.md exists. Do not launch hive workers until role map, blackboard schema, ownership/locks, packet DAG, and verification path are written.

When To Use

Use /hiveworkflow for:

• broad repo audits with independent surfaces
• dense implementation DAGs with clear file ownership
• multi-role planning where scouts, builders, reviewers, and verifiers improve

quality

• high-stakes verification where dissent and independent proof reduce risk
• agent/fleet design where capacity, authority, and safety must be checked from

multiple angles

Downgrade to /workflow when the work is one reasoning loop, one shared file, or one continuous design decision.

Artifact

Create:

mkdir -p .taste/workflow-runs
mkdir -p .taste/hive
STAMP="$(date +%Y%m%d-%H%M%S)"
HIVE_WORKFLOW_ARTIFACT=".taste/workflow-runs/${STAMP}-hiveworkflow.md"
HIVE_RUN_DIR=".taste/hive/${STAMP}"
mkdir -p "$HIVE_RUN_DIR"

Required section order:

# Hive Workflow Run: [task]

## Task
## Taste Gate
## Metacognitive Route
## Hive Eligibility
## Deep Research Brief
## Code Audit
## Role Map
## Blackboard
## Dissent And Conflict Log
## Pre-Plan Introspection
## Plan
## Agent-Native Estimate
## SPEC Decision
## Packet DAG And Ownership
## Execution Notes
## Aggregation Evidence
## Post-Synthesis Introspection
## Verification Evidence
## Outcome

Required Gates

• Capacity: cite bash scripts/parallel-capacity.sh --json.
• Budget: compute effective_hive_budget.
• Role map: every role has purpose, owner, inputs, output, stop condition, and

verification.

• Blackboard: every claim has owner, evidence, status, conflicts, and lock or

merge-barrier state.

• Sidecar: write $HIVE_RUN_DIR/hive-run.json and validate it with

bash scripts/artifact-lint.sh "$HIVE_RUN_DIR/hive-run.json" plus bash scripts/hive-aggregate.sh "$HIVE_RUN_DIR".

• Dissent: non-trivial synthesis has an explicit skeptic/reviewer lane.
• Packet execution: use /parallel sidecars and aggregation when workers edit

files or run implementation packets.

• Introspection: /introspect pre-plan and post-synthesis/post-implementation

hard gates still apply.

• Verification: /verify checks the aggregate result, not hive agreement.

Closeout

Closeout must report:

• hive eligibility decision and any downgrade
• capacity profile and effective hive budget
• roles used and roles intentionally omitted
• blackboard path and claim counts by status
• conflicts, dissent, and arbitration outcome
• packet DAG and aggregation status when used
• verification commands and outcome
• unresolved risks and confidence downgrade

Anti-Patterns

• Launching agents before the blackboard exists.
• Replacing /parallel aggregation with chat synthesis.
• Treating consensus as proof.
• Skipping /introspect because the hive reviewed itself.
• Adding hive workflow pauses unless the user asked for approval.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: waitdeadai
• Source: waitdeadai/minmaxing
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.