Agent Pipeline Development · v6.0.3
CLICK TO COPY /plugin install claude-apd@zstevovich-plugins

Enforced multi-agent pipelines with mechanical guardrails for AI-assisted software development. Available as a plugin for Claude Code and OpenAI Codex.

Interactive Demo → View on GitHub
Last run 2h ago
·
200 tracked · 96 zero-block
·
0 bypass attempts succeeded
Model-Agnostic Enforcement

Same pipeline. Different model. Same result.

APD works as an exoskeleton for any LLM. Here's Claude Opus vs GLM-5 on a TinyMCE → TipTap migration — 7 React components, 99 files.

25×
MORE GUARD BLOCKS ON GLM-5
52 blocks vs Opus's 1–2. Every one a real bypass attempt mechanically stopped — writing to .agents, skipping adversarial, direct code-writes.
100%
PIPELINE COMPLETION · BOTH MODELS
7/7 spec coverage. Full audit trail. The weaker model finished the job because the guards never gave up.
METRIC Claude Opus GLM-5 DELTA
Pipeline time~10m17m 13s1.7×
Completion rate100%100%match
Guard blocks1–25225×
Spec coverage7/77/7match
Builder iterations0–13+
Adversarial findings3–510 (3 real)
Key finding GLM-5 tried to write code directly (blocked by guard-orchestrator), commit without pipeline steps (blocked by pipeline-gate), write .agents via bash (blocked by guard-bash-scope), and skip adversarial review (blocked by verifier). Every bypass attempt was mechanically stopped. Final: 99 files changed, 1349 insertions, 2208 deletions.
The Pipeline

Every step mechanically enforced.

No bypass from within Claude Code or Codex. Each step is blocked by a hook script (CC) or MCP guard tool (Codex) that runs before it completes.

S
Spec
guarded by hooks/spec-hash.sh · max 7 acceptance criteria · hash frozen mid-pipeline
SPECIFY
P
Plan
guarded by hooks/pipeline-gate.sh · plan.done signed before builder can start
PLAN
B
Builder
guarded by hooks/guard-bash-scope.sh · write-scope locked to spec files · orchestrator cannot edit code
IMPLEMENT
R
Reviewer
guarded by hooks/guard-reviewer-order.sh · must run before adversarial · findings logged
REVIEW
A
Adversarial
guarded by hooks/adversarial-gate.sh · context-free audit · zero spec knowledge · opt-out requires spec-card entry
AUDIT
V
Verifier
guarded by hooks/verify-trace.sh · R* criteria traced to test files via @trace · blocks commit if any criterion untested
VERIFY
Commit
guarded by hooks/commit-gate.sh · all 4 .done files HMAC-signed · tamper-proof audit trail
SHIP
Mechanical Enforcement

Every rule backed by a hook or MCP guard.

No honor system. No convention-based hope. CC enforces through hook scripts; Codex through MCP guard tools (apd_guard_write, apd_advance_pipeline). Both exit non-zero when things go wrong.

pipeline-gate.sh

Pipeline Gate

Commits blocked without all 4 pipeline steps. Spec hash frozen mid-pipeline. Max 7 acceptance criteria.

guard-bash-scope.sh

Scope Guards

Agents scoped to specific files. Orchestrator blocked from writing code. Bash writes to pipeline state blocked.

verify-trace.sh

Spec Traceability

R* acceptance criteria traced to test files via @trace markers. Verifier blocks commit if any criterion untested.

adversarial-gate.sh

Adversarial Review

Context-free code review with zero spec knowledge. Finds what code-reviewer misses. Opt-out per task via spec-card.

bin/apd-sign

HMAC-Signed State

Pipeline .done files signed with compiled Go binary. Orchestrator cannot forge completion. Tamper-proof audit trail.

bin/apd-report

Pipeline Report

Full recap in CLI. Step timing, spec coverage bars, adversarial insights, trend analysis, session stats.

apd_mcp_server.py

Codex MCP Server

8 tools — apd_ping, apd_doctor, apd_advance_pipeline, apd_guard_write — bring the same enforcement to Codex. Self-registers via plugin .mcp.json; no per-project install path needed.

plugins/apd/

Self-Contained Plugin

v6.0+ ships every framework binary inside plugins/apd/ — Codex plugin cache contains everything the MCP server needs. Plugin install is the only step. No PATH hacks, no abs-path config, no follow-up `pip install`.

Proven in Production

Numbers that actually mean something.

Every run, every block, every bypass attempt is logged. These stats are live — they come from the same audit trail the verifier uses.

200
logged pipeline runs
across 7 projects
10
guards
that block violations
96/200
zero-block runs
48% clean
0
bypass attempts
that succeeded
Get Started

Three commands. Sixty seconds.

From zero to enforced pipeline.

STEP 01
Add the marketplace
Register the APD plugin source with Claude Code — or with Codex via codex plugin marketplace add zstevovich/claude-apd.
/plugin marketplace add zstevovich/claude-apd Fetching plugin metadata… ✓ Added zstevovich-plugins
STEP 02
Install the plugin
Install APD into your current Claude Code session — or enable in Codex via [plugins."apd@codex-apd"] enabled = true.
/plugin install claude-apd@zstevovich-plugins Installing claude-apd v6.0.3… → 13 hooks → 8 skills (CC) / 7 skills + MCP server (Codex) ✓ Plugin active
STEP 03
Configure and verify
Scaffold your project with the APD pipeline and run the self-check. You should see all 10 guards ready.
/apd-setup Scaffolding .claude/ tree… ✓ .claude/bin/apd, .claude/agents/, .claude/rules/, .claude/memory/ written bash .claude/bin/apd verify ─── APD self-check ─── ✓ pipeline-gate.sh OK ✓ guard-bash-scope.sh OK ✓ verify-trace.sh OK ✓ adversarial-gate.sh OK … 18 more … ✓ all 10 guards ready
Full Walkthrough →