Forward implementation first skill
A skill that keeps an agent off its own hashes, locks and receipts: ship the output first, then verify
- 185 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The skill tells the agent to remove administrative dependencies from the execution path, so the agent edits pipeline scripts and settings.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/forward-implementation-firstDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The skill tells the agent to remove administrative dependencies from the execution path, so the agent edits pipeline scripts and settings.
This is third-party code. Review the repository files before installing.
What it does
Forward implementation first is a 150-line Markdown skill for long pipelines where the agent starts servicing its own paperwork: input hashes, lock files, stage receipts, certification markers, dashboard rows. Before each action the skill makes the agent sort it into one of three classes: implementation, focused validation or administrative bookkeeping. The first two get done, the third is skipped until you ask or it belongs to the product. The forward cursor rule allows a stage to be rolled back or replayed only for a real reason: the input's meaning changed, the target revision changed, the output is corrupted or incompatible with its consumer, a run disproved the earlier result, or the changed producer's dependency cone requires it. If a stage is blocked by a marker alone, the agent runs it by hand, checks the output, publishes, moves on and removes the gate from the path. The skill keeps everything that protects correctness: checksums and signatures users verify, input revision identity, tests that assert results. Parallel workers are allowed under a single writer who publishes results and moves the cursor.
Who it is for. People who run agents on multi-stage pipelines, migrations and long implementation loops and are tired of watching them repair hashes and markers instead of working.
Good fit when
- The agent refuses to run a stage by hand because the pipeline issued no receipt for the previous one.
- After one generator is edited, the agent resets dozens of finished stages over a mismatched hash.
- A migration of a large repository runs for hours without your supervision.
- Several agents or sessions write to one repository, and a single writer has to publish the results.
Not a fit when
- The task is one-off or the session is short and interactive: the author says outright to skip the skill.
- The log and signatures are the deliverable, such as an audit trail or a signed release.
- You need a scheduling policy for parallel workers: the skill has none, your runtime owns it.
Example request
Continue the migration from stage 41 and run it by hand: a mismatched hash on the previous stage is no reason to roll back, just check the output.Limitations
The skill stays a suggestion: the model decides whether to load it, so the rule is duplicated in always-on instructions or agent hooks when needed. Restart the agent after installing, since running sessions do not reload skills. The `install.sh` script copies the file only into the Claude Code and Codex directories and the shared `~/.agents/skills`, so for other agents you place the copy by hand. For a single-project install, copy the directory into the repository's `.claude/skills/` or `.agents/skills/`.
How to disable. Delete the `forward-implementation-first` directory from `~/.claude/skills/`, `~/.codex/skills/` or `~/.agents/skills/`, wherever `install.sh` created it, and restart the agent.
FAQ
Where does install.sh copy the skill?
Into `~/.claude/skills/forward-implementation-first/`, `~/.codex/skills/forward-implementation-first/` and `~/.agents/skills/forward-implementation-first/`, but only those directories it finds on the machine.
What does the skill leave alone?
Checksums and signatures that are part of the product, input revision identity, tests that assert results, benchmarks and end-to-end runs. A missing execution record blocks the claim that relies on it but does not invalidate an unrelated stage twenty steps back.
How do I confirm the agent sees the skill?
Restart the agent and ask it to state the rule for choosing an action. If it cannot, check that `SKILL.md` sits in that agent's skills directory.
What is examples/execution-profile.md for?
It is a starting profile for scheduling parallel workers in case nothing in your setup owns scheduling. The skill never loads it on its own: you copy the file and edit it.
Which pipelines does the author give as examples?
Staged data and ETL runs, large code migrations, build and release pipelines, documentation generation over many inputs, batch analysis and roadmaps an agent works through over days.
Discussion
No questions or reviews yet.
Related
SkillsA completion-discipline skill for large agent tasks: write the acceptance ledger with runnable checks first, then do the work, then reverify against evidence
A skill plus hooks that keep a coding agent on task and cut speculative defenses, unread checksums and scope creep
A workshop skill: the agent vets a finished skill, finds peers, scores it and edits step by step, testing every change.
A meta-skill that folds rules scattered across AGENTS.md, CLAUDE.md and Cursor files into one project skill with task routes.
Self-learning
self-learning-skills
A meta-skill that saves a hard-won working procedure as a new skill or rule, so the next session starts already knowing it
A skill that keeps the task plan in files on disk and re-injects it every turn so work survives context clears, compaction and crashes
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Forward implementation first (this card) | no | 185 | |
| Karpathy-inspired Claude Code guidelines | no | 232,688 | four general principles against LLM coding mistakes; this has one rule on pipeline bookkeeping and stage rollback |
| Planning with Files | no | 28,037 | keeps the task plan in files on disk; this forbids repairing bookkeeping markers in place of output |
README in short
The README walks through a typical failure using a pipeline where one generator change breaks a hash and the agent restarts stages 12 through 40, then covers installation and when the skill fits. The author explains the origin: it began as a prompt pasted at the top of every session. An example run profile sits in `examples/execution-profile.md`.
README badge
Original SKILL.md
--- name: forward-implementation-first description: > Keeps an agent building and validating real output instead of servicing its own bookkeeping. Use for multi-stage pipelines, capability roadmaps, long-running implementation loops, migrations, staged data or build runs, and any run where administrative hashes, locks, receipts, dashboards, certification markers, or progress metadata can block correct work. Use when an agent refuses to advance, rewinds finished stages, or reruns unchanged work because a marker is missing or stale. --- # Forward implementation first Build working capability and correct output before administrative bookkeeping. Apply this contract both while building a system and while running it. ## Scope This skill applies to: - infrastructure and roadmap implementation; - input intake and intermediate artifact production; - transformation, derivation, and analysis stages; - test construction, runtime verification, and measurement; - publication of final output and closeout; - manual invocation of any authorized pipeline stage. It applies whether the action runs through a full orchestrator, a focused command, or a manual stage invocation. ## Decision rule Before each action, classify it as one of: 1. **Semantic implementation**: builds or connects a producer, consumer, adapter, runtime path, schema, fixture, or final output. 2. **Focused validation**: tests the changed dependency cone through behavior, schema, counts, samples, conservation, consistency, nontruncation, or measured resources. 3. **Administrative bookkeeping**: generates or repairs hashes, locks, receipts, dashboards, certification markers, progress metadata, or presence-only records. Choose categories 1 and 2. Skip category 3 unless the user asks for it or the artifact is itself part of the product. When administrative work blocks a path without protecting correctness, remove that dependency from the path. ## Hard constraints