Summary
Record the quality effect and cost trade-off of implementing issue #25 directly versus running the SDD mission loop.
Comparison
| Comparison |
What it measures |
Marketplace diff |
8dfffbd → d1785f5 |
Agent-built feature without the SDD mission loop, including normal follow-up fixes |
6 files, +608 / −0 |
8dfffbd → 4c6ff88 |
Final feature after the SDD mission loop |
6 files, +705 / −0 |
d1785f5 → 4c6ff88 |
Increment uniquely attributable to the SDD loop |
3 files, +122 / −25 |
The SDD-specific increment is a net 97 lines, split as follows:
- Production code: 2 files, +48 / −20; net +28.
- Tests: 1 file, +74 / −5; net +69.
It did not require a different feature architecture or additional production modules; the production increment is concentrated in bounded traversal and containment logic, while most added volume is regression evidence.
Quality improvements
- Bound discovery to direct-child plugin roots and exclude vendor manifest directories, preventing unintended nested or vendor-owned manifests from becoming catalog entries.
- Reject scan roots, plugin directories, manifests, output parents, and output files that resolve outside the repository through symbolic links.
- Replace an unachievable cross-artifact rollback promise with an honest, testable contract: each artifact write is atomic; a later failure is reported with a non-zero exit and does not claim rollback.
- Add regression coverage for nested manifests, vendor directories, all relevant containment boundaries, and a later selected-write failure.
- Produce a frozen 37-scenario suite plus cold spec and implementation reviews, so conformance is based on an explicit contract rather than inherited implementation tests.
Cost trade-off
The SDD mission loop consumed substantially more agent tokens and elapsed work than direct implementation: it required spec re-derivation, control-flow and scenario mapping, iterative cold reviews, gate validation, and durable provenance. No exact token accounting was captured for this mission.
For this change, that extra cost bought concrete quality improvements and corrected two material contract/implementation defects: recursive discovery and the false rollback guarantee.
Summary
Record the quality effect and cost trade-off of implementing issue #25 directly versus running the SDD mission loop.
Comparison
8dfffbd→d1785f58dfffbd→4c6ff88d1785f5→4c6ff88The SDD-specific increment is a net 97 lines, split as follows:
It did not require a different feature architecture or additional production modules; the production increment is concentrated in bounded traversal and containment logic, while most added volume is regression evidence.
Quality improvements
Cost trade-off
The SDD mission loop consumed substantially more agent tokens and elapsed work than direct implementation: it required spec re-derivation, control-flow and scenario mapping, iterative cold reviews, gate validation, and durable provenance. No exact token accounting was captured for this mission.
For this change, that extra cost bought concrete quality improvements and corrected two material contract/implementation defects: recursive discovery and the false rollback guarantee.