ADR-0004: Topic selection¶
- Status: Accepted (2026-09-14); the reinstated group confirms it at the 2026-09-16 kickoff (ADR-0006)
- Date: 2026-09-14
- Deciders: Jerad Dunne, on 2026-09-14 while one member held every role (ADR-0005); confirmed by the four-person group at the kickoff (ADR-0006)
- Decision issue: #41 (proposal P-01)
- Affects research question / method / scope: yes
Context¶
ADR-0002 planned a ranked-ballot vote over proposal issues. When the project briefly became a solo project on 2026-09-14 (ADR-0005), the vote was retired and Jerad Dunne chose the topic directly. On 2026-09-15 the four-person group was reinstated (ADR-0006). P-01 remains selected, and any member may open a Decision needed issue to revisit it before Sprint 1 planning on Thu 2026-09-17.
Options considered¶
Only P-01 was considered, because no other proposals were submitted before the topic was selected on 2026-09-14.
| Proposal | Title | Proposer | Rubric base | Notes |
|---|---|---|---|---|
| P-01 (#41) | Risky capabilities in copied agent skills: prevalence, reach, and drift | Jerad Dunne | Question 4 (skill security and supply-chain risk) with the similarity method from question 1 | Pilot already runs end to end on the GitSkills sample |
Decision¶
P-01 is selected. The research question, as stated in docs/proposals/P-01-jerad-dunne-skill-risk-propagation.md section 2:
In the GitSkills July 2026 sample, how prevalent are skill instructions and bundled scripts that enable risk-relevant capabilities, do skills carrying them reach more repositories through verbatim copying, and do modified variants of the same skill add or remove those capabilities?
RESEARCH_QUESTION.md is completed from the proposal in Sprint 1.
Consequences¶
- Positive: the data fit and the rubric fit are already demonstrated by the pilot; the scanner, rule file, and tests exist.
- Negative / risks: keyword signals may have low precision; RQ3 (variant drift) is thin in the sample; validation depends on agreement between the primary rater and a teammate second rater (ADR-0006). These risks are carried into
THREATS_TO_VALIDITY.mdanddocs/lean-six-sigma/FMEA_RISK_REGISTER.md. - Fallback: rubric question 2 (skill quality indicators), reusing the scanner output as quality features.
- Follow-up issues: complete
RESEARCH_QUESTION.md(#12, #13); decide sample only versus a full-dataset pass by ADR in Sprint 1; build the validation kit (#23).
Revisit trigger¶
A member opens a Decision needed issue before Sprint 1 planning on Thu 2026-09-17, or the Sprint 1 acceptance criterion "the question is answerable with the selected data" is not met by the second check-in on Thu 2026-10-01.