← 2.1.0 Workflow tested · runtime-test

Rapid Skill Development Loop

hot-reload + fork: v1 to v2 with no restart, both on Haiku; fork gives a clean context each run.

Hot-reload with context fork creates a fearless iteration loop for skill development: each test run gets a clean fork context, so failed experiments never pollute the session.

How It Works

The workflow combines two Claude Code 2.1.0 runtime features: hot-reload and context:fork. When a skill declares context: fork, each invocation runs in an isolated execution context. Simultaneously, hot-reload watches the skill's source file on disk. When you edit the skill definition, the next invocation picks up the changes instantly, with no restart required.

This creates a tight feedback cycle: (1) create skill with context: fork, (2) test it—results return clean, (3) modify the skill source in your editor, (4) invoke again—previous context is gone, (5) repeat until the skill works correctly.

What the Test Found

The validation tested skill iteration from v1 to v2 on Haiku. Both versions executed without restart. Fork isolation held: each test run started fresh context, no state leakage from failed experiments. The tester could modify the skill source between runs and observe new behavior immediately on next invocation.

The cleanup is complete. If a skill writes to stdout, modifies files, or leaves data in memory during a failed test, the fork discards it. The next iteration starts truly clean, which eliminates the friction of manual state cleanup or session restart.

Why It Matters

Skill development normally requires careful manual cleanup between test iterations. Either you restart your session each time (expensive), or you manage state manually (error-prone). This workflow eliminates both: every test run is isolated by design, and every code change is live immediately.

The pattern scales to rapid prototyping of agent behaviors, testing hypothesis chains in sequence, and safely exploring edge cases without crashing the main session. Once the skill is stable, you can remove the context: fork flag if you want it to run inline and share session state.

Caveats

Hooks do not fire in forked skills; they execute only in inline skills. For pipeline workflows where skill A needs to trigger skill B via hooks, all skills must run inline. Forked skills are best suited to isolated, self-contained tasks.

Evidence & receipt
  • file2.1.0/WORKFLOW-IDEAS.md
◇ ed25519 receipt
idworkflow_06319270d86a039053f48ad8
alged25519
pubkey9b87705613b1e2fd064d57fa75a6b679d2856ceafad6b1daa8f982493871b6dd
sig04a5f0ef60ffed22fc9a0b9319401efc8ee3d7308789f9012e91d7dce7138f56085514157540b38011b03af60bf4773c46f66b198b8e41ac446b87b0cdf06b0e

Signed with an ed25519 key held off the repo. Anyone can verify against the published public key; nobody without the secret key can forge it. Click verify: it recomputes the signature in your browser. The signature proves integrity and authorship of this exact content — not a third-party timestamp or that the underlying claim is objectively true. signedAt is when the @f3/attest pipeline ran, not when the work happened; the evidence refs carry the source dates.