| name | sdc-harness |
| description | Use when project-level AI guardrails must be generated from SDC standards. |
Codex/Hermes workflow skill generated from commands/harness.md.
Treat the user's current request as $ARGUMENTS. In Codex/Hermes, route to the matching SDC skill/workflow instead of expecting /sdc:* slash-command support.
SDC Harness
Use "$ARGUMENTS" as extra context. Generate or update project-level AI guardrails from .sdc/standards/, confirmed .sdc/knowledge/, and repeated issues recorded in .sdc/memory/.
Run the SDC harness workflow:
- Read
.sdc/constitution.md, .sdc/common-ground.md, .sdc/expert-routing.md, .sdc/standards/, .sdc/knowledge/index.md, relevant confirmed knowledge, project structure, validation commands, historical reports, and user-provided mistakes or constraints.
- If
.sdc/standards/company/README.md exists, read that index first and pull only durable, relevant company/team rules into AGENTS.md.
- Generate concrete
AGENTS.md rules for must-do, must-not-do, validation commands, and known recurring mistakes.
- Convert repeated Common Ground or expert routing misses into concrete guardrails only after user confirmation.
- Preserve
.sdc/constitution.md as higher authority.
- Do not promote memory candidates into
AGENTS.md unless the user confirms they are durable rules.
- Do not invent project rules without evidence or explicit user direction.