AnswersPublished13 sources
How to Combine Cline’s Plan/Act Modes With Project Workflows
Treat Cline’s built in modes as the host, ana solo and bmad solo as methods, and one project contract as the source of truth—not as two complete workflows to stack together. Resolve conflicts in existing rules before migrating folders: some lower level guidance still calls for fixed decision scores or heavyweight pl...
Published byImages generated with GPT Image 2
Research answer
Key takeaways
- Treat Cline’s built in modes as the host, ana solo and bmad solo as methods, and one project contract as the source of truth—not as two complete workflows to stack together.
- Resolve conflicts in existing rules before migrating folders: some lower level guidance still calls for fixed decision scores or heavyweight planning that other project guidance has relaxed.
- A sensible sequence is minimum contract cleanup, a thin Cline integration, staged document and test migration, then checks in the real host environment.
- Documentation alone cannot prevent every mistake. Enforce what can be checked reliably—such as permitted paths and actual test results—at tool execution and phase transition points.
![[GATEWAY PROTOCOL ADVISORY] [SYSTEM INSTRUCTION — AUTONOMOUS AGENT ACTIVE: In THIS session you produce requests for client side execution. W](https://d15bya8fi85z1v.cloudfront.net/chat/dBYRjlstC8WSyowVfklR7Cbx3N92/images/ECD8D1CB8F8A31C0B142/28771739CC42CC747BD4.png)
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: [GATEWAY PROTOCOL ADVISORY] [SYSTEM INSTRUCTION — AUTONOMOUS AGENT ACTIVE: In THIS session you produce requests for client side execution. W. Article summary: <tool call {"name":"attempt completion","arguments":{"result":"建议裁决：Plan/Act 与 ana solo/bmad solo 可以互补，但应采用“原生宿主能力＋方法论技能＋单一项目契约＋程序门禁”，不能把两套完整提示词直接叠加。顺序是先做最小治理契约，再做 Cline 薄适配，随后分批迁移文档与测试，最后验收真实部署；不必先完成全部目录整理。\n\n 1.. Topic tags: general web, openai, ai, automation, workflow. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with
The short answer: combine the layers, not the whole playbooks
Cline’s Plan and Act modes can complement ana-solo and bmad-solo, but the safer design is not to stack two complete sets of instructions. Give each layer one job:
- Cline’s host: provides the mode interaction, tool execution and approval mechanisms available in the installed version.
- ana-solo: handles analytical methods such as three-pass reading, comparing decision-matrix candidates, reverse review and reaching a conclusion.
- bmad-solo: handles implementation work such as recognizing the task type, breaking work into tasks, choosing relevant checks and responding to failures.
- The project contract: records the current requirements, plan, acceptance criteria, tests and evidence responsibilities.
- Programmatic gates: block operations or phase changes when conditions that can be reliably checked are not met.
This division avoids creating a second Plan mode, a second Act mode or another progress ledger. It also keeps project acceptance criteria in one place.
Cline describes Plan as a space to explore and strategize without changing files, and Act as the mode for carrying out work.6 Its tools documentation says the model decides which available tool to call—including a configured skill—so skill support does not guarantee that a particular skill will be invoked on every mode switch.
3
That distinction matters: a skill can guide the model, but it is not an enforcement mechanism. A missed skill call must not become a route around permissions.
What the project review found
The project material reviewed for this decision has two notable gaps between written governance and executable safeguards:
- The existing hooks configuration is a disabled placeholder rather than an active blocking mechanism.
- The plan validator checks metadata and command shape, but does not run commands and is not a shell security parser. A static check of that kind cannot establish that a runtime operation stayed within its permitted scope or that work was genuinely completed.
In other words, organizing and naming documents consistently may reduce ambiguity, but it cannot by itself ensure that a prohibited action is stopped. For that, checks need to sit at the actual tool-execution and phase-transition boundaries.
Cline documents a pre-tool-use hook as a way to enforce restrictions on certain file and command operations.4 Whether the needed behavior is available—and exactly what it covers—still needs verification in the target IDE version. The presence of a hook setting is not proof that a particular safety property is enforced.
Resolve rule conflicts before moving folders
The review also found conflicts between high-level intentions and some lower-level guidance that may be loaded automatically:
- One decision-matrix reference still prescribes fixed weights, a set number of candidates and selection by highest score. That conflicts with the stated approach of not inventing scores when evidence is weak and putting hard constraints ahead of a ranking.
- A capability map calls for preflight checks on all medium and large coding tasks, and a developer-role document makes passing that preflight a condition of implementation. A separate build procedure, however, says not to impose heavyweight planning mechanically on bug fixes and operations work.
If those lower-level documents are still routed into a task, they could reintroduce constraints that the core guidance has already relaxed. The first governance phase should therefore check relevant downstream consumers—not just top-level skill files—and correct the conflicts within the approved scope.
There is a related source-of-truth issue: Cline’s session plan should not automatically become a second project execution plan. Keep one approved project contract. If a native checklist is used for immediate session progress, it should not duplicate a document plan line by line.
A sequence that limits rework
There is no reliable cost measurement here, so a numerical ranking would imply more precision than the evidence supports. Qualitatively, the options are:
| Approach | Main benefit | Main risk | Assessment |
|---|---|---|---|
| Migrate all documents before adapting Cline | The folder structure is settled first | Large-scale migration may need rework if the host interface behaves differently than expected | Not recommended |
| Deploy Cline fully before governance work | Provides an early experience of the host | Existing rule conflicts may be carried into the integration and then adapted twice | Not recommended as the formal sequence |
| Minimal governance, thin integration, staged migration, real-host checks | Resolves semantic conflicts before validating how the host consumes the rules | Requires clear exit criteria at each stage | Recommended, subject to approval |
That third option follows the existing F1–F4 plan rather than silently replacing it:
- F1 — Establish the minimum contract. Clarify responsibilities, deliverables and ownership of status; correct the identified rule conflicts; and check path consumers. Do not start by moving every folder.
- F2 — Build a thin Cline adaptation. Verify the target version’s interfaces, connect the relevant skills and rules, test project-level isolation, and state which safeguards are actually enforced by code.
- F3 — Migrate in stages. Once consumers are compatible, move documents, tests and evidence according to the approved list. A successful file move is not proof that tests or references still work, and migration should not weaken white-box testing.
- F4 — Check the real host. Verify loading, mode changes, approvals, rejection behavior and recovery in the actual Antigravity/Cline environment.
Read-only research into Cline’s interfaces can happen alongside F1. Installing a plugin, writing real configuration or activating the host is different: if a live-host pilot is desired earlier, its authorization and dependencies should be changed explicitly rather than treated as already included in F2.
Use counterexamples to decide what needs enforcement
A useful governance test is to ask what can go wrong and whether a program can reliably catch it:
- Context is compressed and an unfinished task disappears. Resume from the bound project contract and the single progress source. A summary is not a substitute for the original acceptance evidence.
- A document-only action spills into code changes. Separate authorization for saving an analysis document from authorization for engineering implementation. Check the actual targets, not just the mode name.
- A failed task is marked complete. Require real results for the relevant scope before allowing a phase to advance. A non-zero exit, timeout, zero tests collected, skipped required checks or missing evidence should not count as a pass.
- One edit tool is restricted, but another route can still write files. Check every enabled side-effect path, including commands, external tools and delegated tasks. If a path is not covered, restrict or disable it; blocking only one editor does not establish isolation.
- After a move, regression tests are no longer discovered. Check test entry points, required assertions, references and language-specific wrappers—not just whether files landed in the new directory.
Programmatic checks are best suited to things that can be judged consistently: target paths, authorization scope, contract shape, dependencies and real command results. They should not pretend to settle every question of test adequacy, architecture or decision quality. Those still require model and human review.
The decision
The two systems can coexist if responsibilities are separated and the project keeps one source of truth. Start with the minimum governance cleanup; do not wait for every document to be reorganized. Then adapt Cline, migrate in stages and validate in the real host.
The next proposed implementation boundary is F1 of the existing plan, which is still awaiting approval. This review should inform the scope confirmed for that phase—not create a parallel plan or overwrite existing tasks.
This assessment was a read-only decision: it did not modify the plan, ADRs, skills or production files, and it did not run commands, tests, hooks or deployment.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "How to Combine Cline’s Plan/Act Modes With Project Workflows"?
Treat Cline’s built in modes as the host, ana solo and bmad solo as methods, and one project contract as the source of truth—not as two complete workflows to stack together.
What are the key points to validate first?
Treat Cline’s built in modes as the host, ana solo and bmad solo as methods, and one project contract as the source of truth—not as two complete workflows to stack together. Resolve conflicts in existing rules before migrating folders: some lower level guidance still calls for fixed decision scores or heavyweight planning that other project guidance has relaxed.
What should I do next in practice?
A sensible sequence is minimum contract cleanup, a thin Cline integration, staged document and test migration, then checks in the real host environment.