sdd: signal conceptual SDD's process customization is currently bifurcated: standing ...

Author: hlubek

.sdd/graph/2026/05/06-150944-s-cpt-k8i.md

@@ -0,0 +1,17 @@
+---
+type: signal
+layer: conceptual
+kind: gap
+refs:
+    - 20260420-170144-s-prc-p6q
+    - 20260505-171145-s-tac-bv9
+    - 20260410-151529-s-stg-3vr
+participants:
+    - Christopher
+    - Claude
+confidence: low
+summary: SDD's process customization is split between graph-resident contracts and code-bundled extensions, blocking per-project process shaping through dialogue; this signal proposes making moves, hooks, and rules graph-resident via structural injection using Claude Code seams to pull CLI-emitted extensions at session start. It directly addresses the missing "mine" move formalization (s-prc-p6q), builds on the modularization layer now live (s-tac-bv9), and tests whether structural injection overcomes text-instruction unreliability (s-stg-3vr). Open questions remain on which extension levels need new graph kinds, how seams interact with modularization, and whether injection improves behavior control reliability.
+summary_hash: 7cec0d4bf75b29ec4c40e29c67b78863
+---
+
+SDD's process customization is currently bifurcated: standing constraints flow through the graph as contracts (capturable, supersedable, traversable), but playbook moves, rule hooks, and pre-flight extensions ship as code edits to the bundled skill source — preventing per-project process shaping through dialogue. The idea is to keep the base process generic and extend it on demand via graph-resident moves, hooks, and rules, operating at multiple levels: extending pre-flight validation (where contracts already partially live), extending skill instructions, and extending the modular references introduced by the recent modularization (s-tac-bv9). One concrete primitive worth exploring is Claude Code's ability to insert command output at known points in skill markdown — the generic skill could carry seams that pull project-specific extensions emitted by the `sdd` CLI from the graph at session-start, possibly via dedicated entry kinds or repurposed contracts. This complements the existing extension need for the "mine" move (s-prc-p6q), which today would require a code change. Open questions: which extension levels warrant new graph kinds versus reusing contracts; how seams interact with the modularization layer; whether structural injection improves on the text-instruction reliability concern (s-stg-3vr). Captured at idea stage to seed dialogue toward a directive.