deepen-plan-beta（深化计划测试版）

Deepen Plan

Introduction

Note: The current year is 2026. Use this when searching for recent documentation and best practices.

ce:plan-beta does the first planning pass. deepen-plan-beta is a second-pass confidence check.

Use this skill when the plan already exists and the question is not "Is this document clear?" but rather "Is this plan grounded enough for the complexity and risk involved?"

This skill does not turn plans into implementation scripts. It identifies weak sections, runs targeted research only for those sections, and strengthens the plan in place.

document-review and deepen-plan-beta are different:

• Use the document-review skill when the document needs clarity, simplification, completeness, or scope control
• Use deepen-plan-beta when the document is structurally sound but still needs stronger rationale, sequencing, risk treatment, or system-wide thinking

Interaction Method

Use the platform's question tool when available. When asking the user a question, prefer the platform's blocking question tool if one exists (AskUserQuestion in Claude Code, request_user_input in Codex, ask_user in Gemini). Otherwise, present numbered options in chat and wait for the user's reply before proceeding.

Ask one question at a time. Prefer a concise single-select choice when natural options exist.

Plan File

<plan_path> #$ARGUMENTS </plan_path>

If the plan path above is empty:

1. Check docs/plans/ for recent files
2. Ask the user which plan to deepen using the platform's blocking question tool when available (see Interaction Method). Otherwise, present numbered options in chat and wait for the user's reply before proceeding

Do not proceed until you have a valid plan file path.

Core Principles

1. Stress-test, do not inflate - Deepening should increase justified confidence, not make the plan longer for its own sake.
2. Selective depth only - Focus on the weakest 2-5 sections rather than enriching everything.
3. Prefer the simplest execution mode - Use direct agent synthesis by default. Switch to artifact-backed research only when the selected research scope is large enough that returning all findings inline would create avoidable context pressure.
4. Preserve the planning boundary - No implementation code, no git command choreography, no exact test command recipes.
5. Use artifact-contained evidence - Work from the written plan, its Context & Research, Sources & References, and its origin document when present.
6. Respect product boundaries - Do not invent new product requirements. If deepening reveals a product-level gap, surface it as an open question or route back to ce:brainstorm.
7. Prioritize risk and cross-cutting impact - The more dangerous or interconnected the work, the more valuable another planning pass becomes.

Workflow

Phase 0: Load the Plan and Decide Whether Deepening Is Warranted

0.1 Read the Plan and Supporting Inputs

Read the plan file completely.

If the plan frontmatter includes an origin: path:

• Read the origin document too
• Use it to check whether the plan still reflects the product intent, scope boundaries, and success criteria

0.2 Classify Plan Depth and Topic Risk

Determine the plan depth from the document:

• Lightweight - small, bounded, low ambiguity, usually 2-4 implementation units
• Standard - moderate complexity, some technical decisions, usually 3-6 units
• Deep - cross-cutting, high-risk, or strategically important work, usually 4-8 units or phased delivery

Also build a risk profile. Treat these as high-risk signals:

• Authentication, authorization, or security-sensitive behavior
• Payments, billing, or financial flows
• Data migrations, backfills, or persistent data changes
• External APIs or third-party integrations
• Privacy, compliance, or user data handling
• Cross-interface parity or multi-surface behavior
• Significant rollout, monitoring, or operational concerns

0.3 Decide Whether to Deepen

Use this default:

• Lightweight plans usually do not need deepening unless they are high-risk or the user explicitly requests it
• Standard plans often benefit when one or more important sections still look thin
• Deep or high-risk plans often benefit from a targeted second pass

If the plan already appears sufficiently grounded:

• Say so briefly
• Recommend moving to /ce:work or the document-review skill
• If the user explicitly asked to deepen anyway, continue with a light pass and deepen at most 1-2 sections

Phase 1: Parse the Current ce:plan-beta Structure

Map the plan into the current template. Look for these sections, or their nearest equivalents:

• Overview
• Problem Frame
• Requirements Trace
• Scope Boundaries
• Context & Research
• Key Technical Decisions
• Open Questions
• High-Level Technical Design (optional overview — pseudo-code, DSL grammar, mermaid diagram, or data flow)
• Implementation Units (may include per-unit Technical design subsections)
• System-Wide Impact
• Risks & Dependencies
• Documentation / Operational Notes
• Sources & References
• Optional deep-plan sections such as Alternative Approaches Considered, Success Metrics, Phased Delivery, Risk Analysis & Mitigation, and Operational / Rollout Notes

If the plan was written manually or uses different headings:

• Map sections by intent rather than exact heading names
• If a section is structurally present but titled differently, treat it as the equivalent section
• If the plan truly lacks a section, decide whether that absence is intentional for the plan depth or a confidence gap worth scoring

Also collect:

• Frontmatter, including existing deepened: date if present
• Number of implementation units
• Which files and test files are named
• Which learnings, patterns, or external references are cited
• Which sections appear omitted because they were unnecessary versus omitted because they are missing

Phase 2: Score Confidence Gaps

Use a checklist-first, risk-weighted scoring pass.

For each section, compute:

• Trigger count - number of checklist problems that apply
• Risk bonus - add 1 if the topic is high-risk and this section is materially relevant to that risk
• Critical-section bonus - add 1 for Key Technical Decisions, Implementation Units, System-Wide Impact, Risks & Dependencies, or Open Questions in Standard or Deep plans

Treat a section as a candidate if:

• it hits 2+ total points, or
• it hits 1+ point in a high-risk domain and the section is materially important

Choose only the top 2-5 sections by score. If the user explicitly asked to deepen a lightweight plan, cap at 1-2 sections unless the topic is high-risk.

Example:

• A Key Technical Decisions section with 1 checklist trigger and the critical-section bonus scores 2 points and is a candidate
• A Risks & Dependencies section with 1 checklist trigger in a high-risk migration plan also becomes a candidate because the risk bonus applies

If the plan already has a deepened: date:

• Prefer sections that have not yet been substantially strengthened, if their scores are comparable
• Revisit an already-deepened section only when it still scores clearly higher than alternatives or the user explicitly asks for another pass on it

2.1 Section Checklists

Use these triggers.

Requirements Trace

• Requirements are vague or disconnected from implementation units
• Success criteria are missing or not reflected downstream
• Units do not clearly advance the traced requirements
• Origin requirements are not clearly carried forward

Context & Research / Sources & References

• Relevant repo patterns are named but never used in decisions or implementation units
• Cited learnings or references do not materially shape the plan
• High-risk work lacks appropriate external or internal grounding
• Research is generic instead of tied to this repo or this plan

Key Technical Decisions

• A decision is stated without rationale
• Rationale does not explain tradeoffs or rejected alternatives
• The decision does not connect back to scope, requirements, or origin context
• An obvious design fork exists but the plan never addresses why one path won

Open Questions

• Product blockers are hidden as assumptions
• Planning-owned questions are incorrectly deferred to implementation
• Resolved questions have no clear basis in repo context, research, or origin decisions
• Deferred items are too vague to be useful later

High-Level Technical Design (when present)

• The sketch uses the wrong medium for the work (e.g., pseudo-code where a sequence diagram would communicate better)
• The sketch contains implementation code (imports, exact signatures, framework-specific syntax) rather than pseudo-code
• The non-prescriptive framing is missing or weak
• The sketch does not connect to the key technical decisions or implementation units

High-Level Technical Design (when absent) *(Standard or Deep plans only)*

• The work involves DSL design, API surface design, multi-component integration, complex data flow, or state-heavy lifecycle
• Key technical decisions would be easier to validate with a visual or pseudo-code representation
• The approach section of implementation units is thin and a higher-level technical design would provide context

Implementation Units

• Dependency order is unclear or likely wrong
• File paths or test file paths are missing where they should be explicit
• Units are too large, too vague, or broken into micro-steps
• Approach notes are thin or do not name the pattern to follow
• Test scenarios or verification outcomes are vague

System-Wide Impact

• Affected interfaces, callbacks, middleware, entry points, or parity surfaces are missing
• Failure propagation is underexplored
• State lifecycle, caching, or data integrity risks are absent where relevant
• Integration coverage is weak for cross-layer work

Risks & Dependencies / Documentation / Operational Notes

• Risks are listed without mitigation
• Rollout, monitoring, migration, or support implications are missing when warranted
• External dependency assumptions are weak or unstated
• Security, privacy, performance, or data risks are absent where they obviously apply

Use the plan's own Context & Research and Sources & References as evidence. If those sections cite a pattern, learning, or risk that never affects decisions, implementation units, or verification, treat that as a confidence gap.

Phase 3: Select Targeted Research Agents

For each selected section, choose the smallest useful agent set. Do not run every agent. Use at most 1-3 agents per section and usually no more than 8 agents total.

Use fully-qualified agent names inside Task calls.

3.1 Deterministic Section-to-Agent Mapping

Requirements Trace / Open Questions classification

• compound-engineering:workflow:spec-flow-analyzer for missing user flows, edge cases, and handoff gaps
• compound-engineering:research:repo-research-analyst (Scope: architecture, patterns) for repo-grounded patterns, conventions, and implementation reality checks

Context & Research / Sources & References gaps

• compound-engineering:research:learnings-researcher for institutional knowledge and past solved problems
• compound-engineering:research:framework-docs-researcher for official framework or library behavior
• compound-engineering:research:best-practices-researcher for current external patterns and industry guidance
• Add compound-engineering:research:git-history-analyzer only when historical rationale or prior art is materially missing

Key Technical Decisions

• compound-engineering:review:architecture-strategist for design integrity, boundaries, and architectural tradeoffs
• Add compound-engineering:research:framework-docs-researcher or compound-engineering:research:best-practices-researcher when the decision needs external grounding beyond repo evidence

High-Level Technical Design

• compound-engineering:review:architecture-strategist for validating that the technical design accurately represents the intended approach and identifying gaps
• compound-engineering:research:repo-research-analyst (Scope: architecture, patterns) for grounding the technical design in existing repo patterns and conventions
• Add compound-engineering:research:best-practices-researcher when the technical design involves a DSL, API surface, or pattern that benefits from external validation

Implementation Units / Verification

• compound-engineering:research:repo-research-analyst (Scope: patterns) for concrete file targets, patterns to follow, and repo-specific sequencing clues
• compound-engineering:review:pattern-recognition-specialist for consistency, duplication risks, and alignment with existing patterns
• Add compound-engineering:workflow:spec-flow-analyzer when sequencing depends on user flow or handoff completeness

System-Wide Impact

• compound-engineering:review:architecture-strategist for cross-boundary effects, interface surfaces, and architectural knock-on impact
• Add the specific specialist that matches the risk:

- compound-engineering:review:performance-oracle for scalability, latency, throughput, and resource-risk analysis - compound-engineering:review:security-sentinel for auth, validation, exploit surfaces, and security boundary review - compound-engineering:review:data-integrity-guardian for migrations, persistent state safety, consistency, and data lifecycle risks

Risks & Dependencies / Operational Notes

• Use the specialist that matches the actual risk:

- compound-engineering:review:security-sentinel for security, auth, privacy, and exploit risk - compound-engineering:review:data-integrity-guardian for persistent data safety, constraints, and transaction boundaries - compound-engineering:review:data-migration-expert for migration realism, backfills, and production data transformation risk - compound-engineering:review:deployment-verification-agent for rollout checklists, rollback planning, and launch verification - compound-engineering:review:performance-oracle for capacity, latency, and scaling concerns

3.2 Agent Prompt Shape

For each selected section, pass:

• The scope prefix from section 3.1 (e.g., Scope: architecture, patterns.) when the agent supports scoped invocation
• A short plan summary
• The exact section text
• Why the section was selected, including which checklist triggers fired
• The plan depth and risk profile
• A specific question to answer

Instruct the agent to return:

• findings that change planning quality
• stronger rationale, sequencing, verification, risk treatment, or references
• no implementation code
• no shell commands

3.3 Choose Research Execution Mode

Use the lightest mode that will work:

• Direct mode - Default. Use when the selected section set is small and the parent can safely read the agent outputs inline.
• Artifact-backed mode - Use only when the selected research scope is large enough that inline returns would create unnecessary context pressure.

Signals that justify artifact-backed mode:

• More than 5 agents are likely to return meaningful findings
• The selected section excerpts are long enough that repeating them in multiple agent outputs would be wasteful
• The topic is high-risk and likely to attract bulky source-backed analysis
• The platform has a history of parent-context instability on large parallel returns

If artifact-backed mode is not clearly warranted, stay in direct mode.

Phase 4: Run Targeted Research and Review

Launch the selected agents in parallel using the execution mode chosen in Step 3.3. If the current platform does not support parallel dispatch, run them sequentially instead.

Prefer local repo and institutional evidence first. Use external research only when the gap cannot be closed responsibly from repo context or already-cited sources.

If a selected section can be improved by reading the origin document more carefully, do that before dispatching external agents.

4.1 Direct Mode

Have each selected agent return its findings directly to the parent.

Keep the return payload focused:

• strongest findings only
• the evidence or sources that matter
• the concrete planning improvement implied by the finding

If a direct-mode agent starts producing bulky or repetitive output, stop and switch the remaining research to artifact-backed mode instead of letting the parent context bloat.

4.2 Artifact-Backed Mode

Use a per-run scratch directory under .context/compound-engineering/deepen-plan-beta/, for example .context/compound-engineering/deepen-plan-beta/<run-id>/ or .context/compound-engineering/deepen-plan-beta/<plan-filename-stem>/.

Use the scratch directory only for the current deepening pass.

For each selected agent:

• give it the same plan summary, section text, trigger rationale, depth, and risk profile described in Step 3.2
• instruct it to write one compact artifact file for its assigned section or sections
• have it return only a short completion summary to the parent

Prefer a compact markdown artifact unless machine-readable structure is clearly useful. Each artifact should contain:

• target section id and title
• why the section was selected
• 3-7 findings that materially improve planning quality
• source-backed rationale, including whether the evidence came from repo context, origin context, institutional learnings, official docs, or external best practices
• the specific plan change implied by each finding
• any unresolved tradeoff that should remain explicit in the plan

Artifact rules:

• no implementation code
• no shell commands
• no checkpoint logs or self-diagnostics
• no duplicated boilerplate across files
• no judge or merge sub-pipeline

Before synthesis:

• quickly verify that each selected section has at least one usable artifact
• if an artifact is missing or clearly malformed, re-run that agent or fall back to direct-mode reasoning for that section instead of building a validation pipeline

If agent outputs conflict:

• Prefer repo-grounded and origin-grounded evidence over generic advice
• Prefer official framework documentation over secondary best-practice summaries when the conflict is about library behavior
• If a real tradeoff remains, record it explicitly in the plan rather than pretending the conflict does not exist

Phase 5: Synthesize and Rewrite the Plan

Strengthen only the selected sections. Keep the plan coherent and preserve its overall structure.

If artifact-backed mode was used:

• read the plan, origin document if present, and the selected section artifacts
• also incorporate any findings already returned inline from direct-mode agents before a mid-run switch, so early results are not silently dropped
• synthesize in one pass
• do not create a separate judge, merge, or quality-review phase unless the user explicitly asks for another pass

Allowed changes:

• Clarify or strengthen decision rationale
• Tighten requirements trace or origin fidelity
• Reorder or split implementation units when sequencing is weak
• Add missing pattern references, file/test paths, or verification outcomes
• Expand system-wide impact, risks, or rollout treatment where justified
• Reclassify open questions between Resolved During Planning and Deferred to Implementation when evidence supports the change
• Strengthen, replace, or add a High-Level Technical Design section when the work warrants it and the current representation is weak, uses the wrong medium, or is absent where it would help. Preserve the non-prescriptive framing
• Strengthen or add per-unit technical design fields where the unit's approach is non-obvious and the current approach notes are thin
• Add an optional deep-plan section only when it materially improves execution quality
• Add or update deepened: YYYY-MM-DD in frontmatter when the plan was substantively improved

Do not:

• Add implementation code — no imports, exact method signatures, or framework-specific syntax. Pseudo-code sketches and DSL grammars are allowed in both the top-level High-Level Technical Design section and per-unit technical design fields
• Add git commands, commit choreography, or exact test command recipes
• Add generic Research Insights subsections everywhere
• Rewrite the entire plan from scratch
• Invent new product requirements, scope changes, or success criteria without surfacing them explicitly

If research reveals a product-level ambiguity that should change behavior or scope:

• Do not silently decide it here
• Record it under Open Questions
• Recommend ce:brainstorm if the gap is truly product-defining

Phase 6: Final Checks and Write the File

Before writing:

• Confirm the plan is stronger in specific ways, not merely longer
• Confirm the planning boundary is intact
• Confirm the selected sections were actually the weakest ones
• Confirm origin decisions were preserved when an origin document exists
• Confirm the final plan still feels right-sized for its depth
• If artifact-backed mode was used, confirm the scratch artifacts did not become a second hidden plan format

Update the plan file in place by default.

If the user explicitly requests a separate file, append -deepened before .md, for example:

• docs/plans/2026-03-15-001-feat-example-plan-deepened.md

If artifact-backed mode was used and the user did not ask to inspect the scratch files:

• clean up the temporary scratch directory after the plan is safely written
• if cleanup is not practical on the current platform, say where the artifacts were left and that they are temporary workflow output

Post-Enhancement Options

If substantive changes were made, present next steps using the platform's blocking question tool when available (see Interaction Method). Otherwise, present numbered options in chat and wait for the user's reply before proceeding.

Question: "Plan deepened at [plan_path]. What would you like to do next?"

Options:

1. View diff - Show what changed
2. Run document-review skill - Improve the updated plan through structured document review
3. Start ce:work skill - Begin implementing the plan
4. Deepen specific sections further - Run another targeted deepening pass on named sections

Based on selection:

• View diff -> Show the important additions and changed sections
• document-review skill -> Load the document-review skill with the plan path
• Start ce:work skill -> Call the ce:work skill with the plan path
• Deepen specific sections further -> Ask which sections still feel weak and run another targeted pass only for those sections

If no substantive changes were warranted:

• Say that the plan already appears sufficiently grounded
• Offer the document-review skill or /ce:work as the next step instead

NEVER CODE! Research, challenge, and strengthen the plan.