Agent workflow architecture

Codex Feishu Bridge

A working-memory bridge that lets local agent development and cloud-readable project documentation stay in sync.

StatusActive

Year2026

StackFeishu CLI / Markdown / Codex / GitHub

Codex Feishu Bridge

This project is not a visual product. It is an operating loop: local Markdown remains the source of truth, while Feishu becomes the cloud-readable and commentable layer.

#Why it exists

Long agent-led projects fail when state lives only in chat. A useful bridge gives each project a stable map, a current context, an execution log, and a place for human comments.

#System shape

Local repository Markdown stores decisions and phase state.
Feishu Wiki mirrors the key documents for cloud review.
GitHub keeps source, issues, PR templates, and CI.
Codex uses the local map to continue work without restarting from memory.

#What it protects

The bridge protects against three failure modes:

A new agent repeats solved work.
The user cannot comment on project direction in the cloud.
A deployment or audit failure disappears into chat history.

This is a small project, but it is a real external workflow around the homepage rather than another page inside the homepage.

Evidence Pack

Proof you can inspect

Proof #1 / PrimaryFeishu Wiki surface

The cloud-readable knowledge base gives project maps, decisions, progress logs, and phase reviews a commentable surface outside the local repo.

It proves the bridge has a real cloud endpoint for human review, not only local planning files.

Feishu/knowledgewiki.open()2026-07-03

Open evidence Proof #2 / SupportingHandoff contract

The bridge treats chat as exploration and repository Markdown as the durable fact source for future agents.

It explains the governance decision that makes the bridge durable rather than merely convenient.

GitHub/projects/codex-feishu-bridgehandoff.audit()2026-07-03

Open evidence Proof #3 / ContextLocal sync protocol

`docs/FEISHU_SYNC.md` records which local documents map to Feishu nodes and how the sync loop should be maintained.

It captures the repeatable maintenance rule behind the bridge, which is essential for future agents.

GitHub/projects/studio-knowledge-basedocs.trace()2026-07-03

Open evidence

Case Study Diff

What changed

Change 01

Give operating state a named home

Before

Project state was distributed across chat, local files, and memory.

Constraint

The loop had to remain light enough for agent work while leaving a readable, commentable trail for human review.

After

Current state, decisions, and phase progress have named documents and a Feishu cloud mirror.

Change 02

Convert audit output into owned work

Before

Audit findings could remain a one-off report.

Constraint

Findings need priorities and phase ownership, otherwise they become documentation theater.

After

Audit findings become a prioritized todo with P0-P3 ownership and roadmap linkage.