Repeatable workflow packs for turning Codex subagents into auditable deliverables.
codex-workflow-packs is a local Typer CLI for generating structured Codex workflow prompts, runbooks, report scaffolds, and verification checklists. It is for developers, consultants, maintainers, and technical operators who want repeatable Codex-assisted work instead of one-off prompt improvisation.
The problem it solves: a random list of agents says who might help, but it does not define the job. A workflow pack combines the input contract, recommended subagent roles, ordered steps, verification checks, and expected deliverables so the work can be reviewed, repeated, and handed off.
The CLI does not run Codex or custom subagents automatically. It prepares the materials for explicit Codex usage by a human operator.
python3 -m venv .venv
. .venv/bin/activate
python -m pip install -e ".[dev]"
codex-packs list
codex-packs plan repo-audit --input examples/repo-audit-input.md
codex-packs scaffold-run repo-audit --output runs/demoExpected output:
codex-packs listprints the available workflow packs and their deliverables.codex-packs plan ...prints a structured Codex prompt with required inputs, subagent sequence, workflow steps, deliverables, and verification checks.codex-packs scaffold-run ...creates a local run folder with prompt, input, evidence, deliverable, report, and verification files.
See examples/sample-output for small example deliverables.
Input -> workflow pack -> explicit subagent prompt -> verification checklist -> deliverables
Each pack defines:
- required and optional inputs;
- recommended Codex subagent roles to invoke explicitly;
- ordered workflow steps;
- expected output files;
- verification checks for review.
- Not an autonomous agent platform.
- Not a replacement for Codex.
- Not a security, compliance, or legal guarantee.
- Not a magic wrapper that secretly runs all agents.
Generated prompts and reports require human review. This tool helps structure work; it does not guarantee fixes, security outcomes, or production readiness.
| Pack | Use it for | Sample deliverables |
|---|---|---|
repo-audit |
First-pass repository review before planning fixes. | Audit report, risk register, fix backlog, runbook. |
bug-triage |
Disciplined bug analysis before code changes. | Reproduction notes, likely cause, minimal fix plan, regression test plan. |
linux-vps-audit |
Preliminary Linux/VPS incident or hardening review. | Incident report, command plan, risk register, hardening checklist. |
data-cleanup |
Human-reviewed data cleanup and QA workflow. | Cleaned data plan, QA report, transformation log, anomaly list. |
ai-workflow-eval |
Prompt, agent, or workflow regression evaluation. | Eval cases, failure taxonomy, regression report, improvement backlog. |
python3 -m venv .venv
. .venv/bin/activate
python -m pip install -e ".[dev]"Run tests:
pytestList packs:
codex-packs listShow a pack:
codex-packs show repo-auditValidate packs:
codex-packs validateGenerate a plan without running Codex:
codex-packs plan repo-audit --input examples/repo-audit-input.mdCreate a run folder:
codex-packs scaffold-run repo-audit --output runs/demo# Workflow Plan: repo-audit
## Required Inputs
- repository context: Repository path, project purpose, target branch, and any areas of concern.
## Agent Sequence
1. code-mapper
2. dependency-manager
3. test-automator
4. reviewer
5. security-auditor
6. technical-writer
## Deliverables
- repo-audit-report.md
- risk-register.csv
- fix-backlog.md
- runbook.md
- 24h repo audit.
- Bug reproduction report.
- Linux/VPS preliminary incident report.
- Data cleanup + QA report.
- AI workflow regression pack.
See docs/MONETIZATION.md for fixed-scope examples.
Clone or download the agent collection, then preview:
codex-packs install-agents --source ./awesome-codex-subagents --dry-runInstall globally into ~/.codex/agents:
codex-packs install-agents --source ./awesome-codex-subagents --no-dry-runThe installer copies nested .toml files and preserves folder structure. Codex custom subagents still must be explicitly invoked in a Codex session.