Implement Development Tracks

Implement Track is a skill for executing tasks from a Conductor track's implementation plan, following the rules defined in conductor/workflow.md. It first verifies Conductor is initialized (product.md, workflow.md, tracks.md exist), loads the workflow's TDD strictness, commit strategy, and verification-checkpoint rules, then selects a track - either the one named as an argument, or, if none given, a menu of in-progress and pending tracks parsed from tracks.md.

Context loading pulls the track's spec, plan, and metadata, plus project-level context (product understanding, tech stack, workflow rules, and language-specific style guides). Each incomplete task in the plan is worked in a loop: marked in-progress, then either run through TDD (write a failing test, confirm it fails, implement the minimum code to pass, confirm it passes, refactor while keeping tests green - halting if a test passes unexpectedly before implementation) or, if TDD isn't strict, implemented directly with existing tests run and manual verification as needed. Completing a task commits the change with a workflow-defined prefix, marks the task done in plan.md (committed separately), and updates metadata.json's completed-task count. After each phase's tasks are all done, it runs phase verification (the full test suite) and explicitly waits for the user's approval before continuing to the next phase - this checkpoint is never skipped.

Failures halt immediately rather than continuing automatically: a tool failure offers retry/skip/pause/revert, a test failure offers fix/rollback/pause, and a git error (uncommitted external changes, merge conflicts, permissions) offers status/resolve/pause - in every case the skill presents options and waits rather than guessing. When all phases and tasks are complete, it runs final verification against the spec's acceptance criteria, marks the track complete in tracks.md/metadata.json/plan.md, offers to sync documentation (product.md, tech-stack.md, README) and to archive/delete/keep the track directory, and prints a completion summary (phases/tasks/commits/test status). If implementation is paused and resumed later, it reloads metadata.json's current task and asks whether to continue, restart the current task, or show a progress summary first.

Use it to execute a Conductor track's plan task-by-task with enforced TDD and mandatory phase-approval checkpoints, rather than implementing freeform without tracked progress. It is not for projects without Conductor initialized - the pre-flight check explicitly fails and directs you to run setup first rather than improvising a plan structure.

Input is a track ID (or none, to pick from a menu) plus the track's spec/plan/metadata files. Output is committed code changes, an updated plan.md and metadata.json tracking progress and commit hashes, and phase/track completion reports requiring explicit approval to proceed.

It operates on the local conductor/ directory structure and git, running the project's own test suite (npm test, pytest, etc.) as its verification mechanism at each phase and at track completion.

Developers using the Conductor workflow system who want disciplined, TDD-enforced task execution with mandatory human approval between phases, full progress tracking, and safe pause/resume rather than ad hoc implementation.