[sabind]

Why

#63 compiles every agent's plugin from one canonical source. The approved strategy said the opposite, "converge first, compile at the edges", so it is revised before that stream is designed. The spike showed agents differ in exactly the places neutral markers express, one plain-word fallback already failed in Codex, and one folder per agent keeps each agent's format changes in one place.

What

Revises the multi-tool plugin strategy so that every agent's plugin, Claude Code's included, is compiled from one canonical source, and renames its streams of work and their candidate designs to match.

Changes

• Revise .freethinking/strategies/multi-tool-plugin.md:
  • The principle. "One canonical source, compiled for every agent", with one target folder per agent, replaces "converge first, compile at the edges".
  • Goal 3. One canonical source, with a freshness and completeness check in CI.
  • Goal 1's measure. Every shared scenario for skills, sub-agents and rules that passes in Claude Code passes in each tool.
  • The streams. They are now canonical source and the Claude Code target, which must reproduce today's plugin byte for byte; agent targets, adding Devin; and cross-agent evals, with checks along the way, weighing the author's Gherkin harness.
  • The risks. Three new risks: markers growing into a language, a build on every change, and the migration changing Claude Code's behaviour.
  • The design candidates. Renamed to canonical-source, agent-targets and cross-agent-evals.

Testing

Not tested

Risk and rollout

None

🤖 Generated with Claude Code