Always load the Elon method Algorithm order and a small software design core #206

What: The global contract always carries the Elon method's ordered Algorithm and a small software design core; detailed references stay on demand.
Why: The intent pilot measured that Elon method and software design apply on about half of all agent turns and are 80% of moments where a procedure applies, so per-turn detection costs more than carrying the core.
So what: An agent implements the change with a PR and independent review; the user decides merge and release.

Outcome

Every agent session has the Elon method's core rules and the software design core in context without loading operations. Business or non-code work can skip the software design core by its stated condition.

Acceptance criteria

• global/AGENTS.md states the Algorithm's binding order (question, delete, simplify, accelerate, automate) next to the three tests, without copying the Algorithm's completion conditions.
• global/AGENTS.md has a software design section of about 1,000 to 2,500 characters that states when it applies and when to skip it, and points to the procedure for deeper choices.
• workflows/software-design/index.md no longer duplicates rules the core now owns, and the routing row still selects it for deeper design choices.
• Contract tests pin the new core text and keep detailed Algorithm completion conditions out of the core.
• Full test suite passes; the added always-on character count is reported.

Non-goals

• Moving software design references (compare designs, state and boundaries, TypeScript, change existing systems) into the core.
• The conversational Elon record of Elon method is not applied in conversation: give it a visible output, concrete triggers and a brief rule #198 or the core split of Test moving code-delivery rules out of the always-loaded core #199.
• New hooks or skills.

Blockers

None.

Proof

python3 -m unittest discover -s tests, versionctl release-check, character count of global/AGENTS.md before and after, independent review of the PR.

Objective contribution

Makes the procedures that most often apply reliable in ordinary work without loading operations, at a bounded always-on cost.

Elon record

• Requirements and who asked: the user (2026-10-10), from the intent pilot measurement: Elon method always on; software design always on only if its core stays small, because agents also do business tasks where it does not apply.
• Deleted: per-turn detection of these two procedures; copying the full Algorithm or the four design references into the core; any new hook or skill.
• Bottleneck: the rules are reachable only through an operations load that agents often skip; the pilot shows they apply too often to detect cheaply.
• Checked myself: global/AGENTS.md is 16,395 characters and already carries the three tests; workflows/software-design/index.md is 2,557 characters; tests/test_skill_contracts.py currently forbids Question every requirement. in the core, so the ordered steps must be phrased as a compact order, not a copy.