| name | dev-strategy |
| description | Creates sufficient implementation plans for complex coding projects by analyzing your codebase, gathering constraints, and proposing execution strategies. Use when you need to "plan how to build", "create implementation roadmap", "break down this feature", "how should I approach building", or structure multi-phase development, refactoring, or architectural changes. Produces code context, behavioral specs, decision logs, and implementation roadmaps. |
| skills-depend-on | ["contribution-system","design-principles","execution-strategies"] |
Dev Strategy Skill
Transform user requirements into actionable development roadmaps through structured planning and codebase analysis.
What This Skill Does
Creates sufficient implementation plans by:
- Analyzing your codebase architecture and patterns
- Gathering constraints and requirements
- Proposing execution strategies
- Generating structured artifacts for implementation teams
Execution Overview
To produce a sufficient implementation plan, this skill progresses through three outcomes:
Each criterion below is either confirmed by the user or checkable against an artifact. These criteria are owned here; references restate them for execution convenience only.
Outcome 1 (UNDERSTANDING): Context is clear
- Behavioral spec is confirmed by the user
- Draft
code-context.md exists, naming the files/functions the change will touch, with line references
- Constraint list is recorded in
context-document.md and explicitly confirmed by the user
- (Technical research is governed by the trigger conditions in references/tactics.md — there is no standing "research done" criterion)
Outcome 2 (STRATEGY): Approach is selected
- User selected a strategy (consult
execution-strategies skill)
decision-log.yaml records the choice, rationale, and rejected alternatives with one-line reasons
Outcome 3 (PLANNING): Execution is clear
- All 4 core artifacts exist (code-context.md, context-document.md, decision-log.yaml, implementation-roadmap.md) and respect their templates
- Roadmap is phased appropriately (each phase is a complete deliverable)
- First phase lists concrete file paths and testing criteria
- Future phases are deferred appropriately per Last Responsible Moment principle
For the generic outcome pattern shared by all skills, see contribution-system/references/execution-template.md.
Key Outputs
The skill generates only the following structured artifacts in .plans/plan-[FEATURE-NAME]/:
Core Artifacts:
- Code Context - Relevant classes/functions with line references
- Context Document - Behavioral spec + architecture summary + research findings
- Decision Log (YAML) - Technical decisions made during strategy development
- Implementation Roadmap - Phased execution plan with strategy
Research Artifacts (when research tactic is applied):
- Research Summary - Key findings and recommendations
- Technology Research - Focused integration analysis of new technologies
- Implementation Examples - Code patterns and usage examples
Guiding Principles
For principles applied during planning (YAGNI, Last Responsible Moment, Kent Beck's 4 Rules, Sufficient Planning), see design-principles skill.
Token Efficiency — a plan's cost is not just the planning session; every artifact is re-read by every downstream contribution:
- Delegate bulk reading. Codebase analysis and technical research go through sub-agents (Explore for code,
model: "haiku" — bounded retrieval; general-purpose for research, default model — synthesis feeds artifacts every later phase relies on) that return distilled findings — raw file contents and web-search transcripts must not accumulate in the planning session.
- Artifacts are paid for once per phase. A line in
context-document.md or code-context.md is re-read by every contribution, so an N-phase plan pays for it N times. Write artifacts lean: distill, don't transcribe; link to code by path and line range instead of quoting it.
- Size phases for a single fresh session. Each phase must be completable by one dev-contribute invocation in a fresh context window without approaching the context limit (~100k tokens of working room). A phase that needs more is two phases.
- Make phases self-contained in the roadmap. Each phase entry should carry the file paths, constraints, and success criteria the implementer needs, so a contribution can start from a compact brief instead of re-reading the whole plan.
- Interactive replies: lead with the outcome, one level of detail. No process narration, no options not chosen, no teaching material — the user asks if they want more. But keep decisions, trade-offs, and failures: omitting one triggers a follow-up round-trip over the full context, costlier than the sentence that preempts it. Conversation only — artifacts are read by fresh-context agents that can't ask, so they stay complete-but-lean.
How to Use This Skill
For tactics available for each outcome, see references/tactics.md.
See Also
Related Skills:
- execution-strategies — To compare TDD, Steel Thread, and Core-then-Integrate approaches during strategy selection
- dev-contribute — To implement the planned phases once dev-strategy is complete
- design-contribute — When a planned phase has design objectives that need collaborative refinement
- design-principles — To understand YAGNI, Last Responsible Moment, and other principles applied during planning
- contribution-system — For artifact schemas and folder naming conventions
