| name | weave |
| description | Compose and run a large delivery as ONE pipeline with magician's guardrails — for big multi-item work ("implement these N stories/tickets", "deliver the epic", "migrate X across the codebase", batch/sweep). Adaptive structure, but always TDD per unit, kg grounding, certify, multi-lens review + adversarial verify, write gates, no context loss. Use instead of hand-rolling many agents. |
$weave — Codex Adapter
Read ../../references/codex-adapter.md, then read ../../../skills/weave/SKILL.md and follow the source skill through that Codex adapter. Keep the source skill's gates and non-negotiables.
Codex equivalents:
- Engine — if agent tools are available, compose the delivery per
../../../skills/weave/references/template.md: sequential TDD per unit → $certify → parallel multi-lens review → adversarially refute every Critical/High → consolidate. Use generic agents with self-contained role prompts; reviewer, security, simplification, and verification are responsibilities, not magician:* profiles. Otherwise execute the same stages locally.
- Grounding —
kg check/kg init, then kg query/kg blast to scope each unit; pass file:line pointers into worker prompts, never whole files.
- Write gates — implement + test on a branch; never push, open/merge PRs, or do anything destructive without explicit approval. A unit commit requires authorization and must stage only enumerated task-owned files.
- No context loss — every worker prompt self-contained (Goal/Scope/Inputs/Constraints/Return); workers return distilled summaries.
- Plan gate — show units + structure + rough scale and get approval before running a large pipeline (use Codex's question/approval UI).
