Forward Implementation First Agent Skill Guide

Key Takeaways#

• forward-implementation-first is a model-agnostic skill for keeping coding agents focused on product output.
• It targets a specific failure mode: agents servicing hashes, receipts, dashboards, and other bookkeeping instead of shipping the requested artifact.
• The core rule is to do semantic implementation and focused validation, while skipping administrative bookkeeping unless the user asked for it or the artifact is part of the product.
• The repository is short, MIT-licensed, and intended for Claude Code, Codex, and other coding agents.

What this resource covers#

forward-implementation-first is a developer-agent skill for long-running implementation work. Its premise is that agents can become too obedient to their own process artifacts. In a pipeline, they may generate content hashes, lock files, completion receipts, certification markers, dashboards, or progress metadata. Those artifacts can be useful in a real product contract, but they can also become fake work when the actual requested output is code, data, docs, or a working runtime path.

The README describes a staged pipeline where a changed producer makes an old hash mismatch. A cautious agent can then invalidate many stages, refuse to run the next stage manually, and spend hours repairing records that do not change the final product. Each local decision looks defensible. Together, the project stalls.

The skill's decision model#

The skill asks the agent to classify actions before doing them. The three categories are semantic implementation, focused validation, and administrative bookkeeping. Semantic implementation includes building or connecting producers, consumers, adapters, runtime paths, schemas, fixtures, and final outputs. Focused validation includes tests and checks that prove the changed dependency cone still works. Administrative bookkeeping includes repairing hashes, locks, receipts, certification markers, dashboard rows, or presence-only records.

The rule is direct: do the first two, skip the third unless the user requested it or the bookkeeping artifact is part of the product. If a bookkeeping dependency blocks a path without protecting correctness, remove that dependency rather than making it the job.

Where it fits#

This skill is most useful when a coding agent is operating a multi-step content, data, or build pipeline. It also helps in agentic refactors where there are many tempting side tasks. The skill pushes the model toward concrete implementation and measured verification rather than process theater.

Use it with Claude Code, Codex, or any agent environment that accepts project instructions or skills. It is especially relevant for teams that run agents unattended, because unattended agents are more likely to over-serve artifacts that look like proof of progress.

Adoption checklist#

1. Read the official README and skill file.
2. Add the skill to your agent environment.
3. Use it on a workflow where the agent previously got stuck maintaining status files.
4. Watch whether the agent prioritizes changed producers, changed consumers, and real output checks.
5. Keep real integrity mechanisms intact; do not use the skill to skip product-level hashes, signatures, tests, or benchmarks.

Limitations#

forward-implementation-first is not permission to ignore validation. The repository explicitly preserves real evidence, tests, reproductions, and product-contract integrity. Its point is to separate correctness checks from administrative shadows of correctness. If your artifact is the receipt, the receipt matters. If your artifact is working code or usable content, do that first and verify the behavior.

Source#

Official repository: https://github.com/Vuk97/forward-implementation-first