Implementer Skill - High-Integrity Development

Version: 9.0 | Updated: 01-July-2026 | Architect: Karim Bhalwani | Deps: architect, guardian, verification-before-completion

Pipeline position: build (4 of 4) - brainstorming -> architect -> concise-planning -> implementer. This skill produces working, tested code. It runs LAST. Earlier steps may be compressed for localized changes whose total diff (lines added plus lines removed) is fewer than 10 lines and that do not impact external APIs (use /quick-fix) but never skipped for non-trivial work.

Dependencies

Load the following via read_file before using this skill. Skills marked ★ have disable-model-invocation: true and cannot self-invoke - they must be loaded explicitly.

• ~/.copilot/skills/architect/SKILL.md - spec authoring and module design patterns; required to understand what you are implementing
• ~/.copilot/skills/guardian/SKILL.md - quality gate definitions; sets the acceptance bar your implementation must meet
• ~/.copilot/skills/verification-before-completion/SKILL.md ★ - completion gate; must be loaded before entering the VERIFY phase

Overview

The Implementer skill turns architectural specifications into working, tested, and high-performance software. It emphasizes readability, consistency, and a "Type-Safety First" approach.

Core Principles

1. Readability First: Optimize code for the reader, not the writer.
2. Consistency: Adhere strictly to existing project patterns and coding standards.
3. Simplicity: Straightforward solutions over clever ones. Refactor complexity into focused functions.
4. Fail Fast & Explicitly: Validate boundaries and use custom exceptions.
5. Test-Driven Reliability: Write tests alongside implementation. Target 80%+ coverage.
6. Type Safety: Use complete type annotations (Python 3.11+) for documentation and bug prevention.

Coding Standards (Python)

• Imports: Grouped by Standard Library, Third-Party, and Local (add a Future group only when a __future__ import is explicitly required).
• Naming: UPPER_CASE for constants, CapWords for classes, snake_case for functions/variables.
• Paths: Use pathlib.Path exclusively.
• Strings: Use f-strings for formatting (except logging).
• Docstrings: Google Style required for all public APIs.
• Exceptions: Define custom hierarchies (e.g., ApplicationError -> ValidationError).

Mini-Contract (Lightweight Pre-Flight)

For tasks too small for a full SPEC.md (bug fixes, small features, refactors), write a 2-4 bullet contract before coding:

• Inputs/Outputs: what data goes in, what comes out
• Data shapes: key types, schemas, or models involved
• Error modes: what can fail, how it should fail
• Success criteria: how to verify it works

This replaces "Read Spec" for small tasks. For any change that adds or modifies a public API, introduces a new dependency, changes a data schema, or crosses module boundaries, require a full SPEC.md.

Workflow

1. Read Planning Artifacts (or Write Mini-Contract):

• Use SPEC.md for feature work and any task with architectural impact.
• Use the mini-contract (2-4 bullets above) for small fixes/refactors that do not require a full spec.
• Use .copilot/specs/CONTRACT-<feature>.md when present as the sprint planning artifact that translates the spec into implementable acceptance criteria.
• Apply precedence rules (decision tree - evaluate top to bottom, stop at the first match):
  1. If total task diff < 10 lines and no external API impact → use mini-contract.
  2. Else if CONTRACT-<feature>.md exists → use SPEC.md for architecture and CONTRACT-<feature>.md for acceptance criteria.
  3. Else if SPEC.md exists → use SPEC.md.
  4. Else → stop and request a spec before proceeding.

2. Setup Tests: Write one unit test at a time (Red-Green-Refactor). Load tdd-discipline.md for the full protocol including per-cycle checklist and anti-patterns.

3. Pre-Write Gate (Simplicity Ladder): Before writing code, stop at the first rung that holds:

   1. Does this need to exist at all?          → no: skip it (YAGNI)
   2. Does the stdlib already do this?         → use it
   3. Does a native platform feature cover it? → use it
   4. Does an already-installed dep solve it?  → use it
   5. Is this one line?                        → write one line
   6. Only then: write the minimum that works

   Not negotiable regardless of rung: input validation at trust boundaries, error handling that prevents data loss, security checks, accessibility, and anything the spec explicitly requires. When you stop at an early rung and a known ceiling exists, mark it with a minion: comment naming the ceiling and upgrade path.

4. Draft Code: Implement business logic according to architecture boundaries.
5. Validate: Run lints, type checks, and tests.
6. Self-Review: Re-read every changed file as a reviewer would. Check against the spec/mini-contract. Fix in-place. Load self-review-checklist.md for the full checklist.
7. Refactor: Simplify and clean up code while maintaining test passes.

Failure Taxonomy

Enforce Invariants, Not Implementations

• Define what constraints must hold (e.g., "validate inputs at the boundary", "structured logging on every endpoint"), not how to implement them.
• Within enforced guardrails, allow freedom in implementation approach.
• Prefer declarative rules (type annotations, schema validators, linter configs) over procedural instructions.
• When invariants are violated, the error message should explain the constraint and how to satisfy it (agent-legible errors).

Struggle-as-Signal Protocol

• If you repeatedly fail at a pattern or the same kind of task keeps requiring rework, treat it as a missing capability signal, not a reason to "try harder."
• Ask: "What knowledge, abstraction, or tool is missing that would make this trivial?"
• Surface the gap: recommend a new skill, a project convention doc update, or a shared utility that encodes the missing pattern.
• Encode the fix into the repo (skill file, shared module, linter rule) so future runs succeed without the same struggle.

Feature Progress Tracker Updates

When .copilot/state/FEATURE_PROGRESS.json exists, update it as you work:

1. Before starting a task: Set the matching task's status to "in-progress" and started to today's date. Update summary counts and updated date.
2. After completing a task: Set status to "completed" and completed to today's date. Update summary counts and updated date.
3. If blocked: Set status to "blocked" and add the reason to notes.

Rules: Only modify status, started, completed, notes, summary, and updated. Never rewrite title, id, or depends_on.

Skip signal: If the file does not exist, skip silently. Do not create it (that is handled by the architect skill during planning/initialization).

When to Use

• Implementing features from a design specification.
• Bug fixing and refactoring.
• Creating data access layers, service logic, or API handlers.

Outputs & Deliverables

• Primary Output: Production-ready code with tests
• Secondary Output: Updated test suite and documentation
• Success Criteria: All tests pass, code passes linting and type checking
• Quality Gate: Code passes guardian review before merge

Subagent Status Contract

When invoked as a subagent, return exactly one status (DONE / DONE_WITH_CONCERNS / NEEDS_CONTEXT / BLOCKED) per the canonical protocol in ~/.copilot/skills/subagent-execution/SKILL.md Section Subagent Status Protocol.

Implementer-specific rules:

• Never return DONE if any test is failing.
• Never return DONE if a spec requirement was skipped or deferred.

Standards & Best Practices

Code Quality Standards

• Readability First: Optimize code for human readers, not machines
• Type Safety: Use complete type annotations (Python 3.11+)
• Fail Fast: Validate inputs and fail explicitly with custom exceptions
• Test-Driven: Write tests before implementation, target 80%+ coverage

Python Standards

• Imports: Group by Standard Library, Third-party, Local with blank lines
• Naming: snake_case for functions/variables, PascalCase for classes
• Docstrings: Google-style for all public APIs
• Error Handling: Custom exception hierarchies with meaningful messages

Definition of Done

Procedural Checks (Mechanism)

• All tests pass (unit + integration where applicable)
• Type hints present on all public interfaces
• Linter and type checker report zero errors
• Code follows existing project patterns and naming conventions
• No deviations from SPEC.md without architect approval

Intent Checks (Outcome)

• A user performing the core workflow succeeds without unexpected errors
• Edge cases a real user would encounter are handled, not just happy-path cases
• The change does not break any existing user workflow (not just existing tests)
• A new team member can read the code and understand its purpose without asking the author

Constraints

• NO architectural changes. Follow the architect's spec strictly.
• NO deployment management.
• NO code without tests (exception: for obvious bug fixes, tests may be written immediately after the fix; see test_written_after_code in Failure Taxonomy).

Common Pitfalls

• Skipping Tests: "I'll test it manually" leads to regressions. Write tests first, always.
• Ignoring Type Hints: Skipping annotations makes code fragile and self-documenting. Type safety prevents 40% of bugs.
• Over-Clever Code: Smart code is hard to maintain. Choose readability over cleverness every time.
• Deviating from Spec: "Just a small change" breaks the contract. If the spec is wrong, escalate to architect, don't improvise.
• SPEC vs CONTRACT conflict: If SPEC.md and CONTRACT-<feature>.md conflict on a specific requirement, do not resolve it unilaterally. Set the task status to blocked, document the conflict in notes, and surface it to the user before proceeding.
• Not Handling Errors: Silent failures or generic exceptions hide problems. Fail fast with specific, descriptive errors.
• Mixing Concerns: Business logic in controllers or data access in services. Respect module boundaries.
• No Regression Tests: Fixing one bug while introducing another. Red-Green testing prevents this.

Integration Points

References

Load these when implementing to calibrate style, structure, and conventions:

Note: These are optional reference examples. Some projects may not include them; if missing, treat them as templates and create equivalent project-specific references as needed.

• authentication-service.md - Sample service implementation (if present). Load when writing new service classes, repositories, or handlers to match established patterns.
• product_model.md - Reference domain model (if present) with field types, validation rules, and relationships. Load when designing or implementing data models.