Optimize the contributor-growth skill family #1348

Activity

1. potiuk

   added

   enhancementNew feature or requestNew feature or request

   kind:perfToken / latency / API-call budgetToken / latency / API-call budget

   family:contributor-growthcontributor-growth skillscontributor-growth skills

   on Sep 22, 2026
2. potiuk

   added a parent issue on Sep 22, 2026

   • Optimize skill token cost across all families (setup-family recipe) #1342
3. Kaap10 commented on Oct 1, 2026

   

   Kaap10

   on Oct 1, 2026

   Contributor

   More actions

   Hi @potiuk

   I'd like to pick up #1348 and optimize the contributor-growth skill family following the #1342 recipe.

   To keep reviews atomic and bite-sized, I plan to break the work into focused PRs:

   Phase 1: Always-On Frontmatter Trimming (Priority 1 — Target $\le 200$ tokens)

   • PR 1: onboarding-concierge (~275 tokens ⚠ $\rightarrow \le 200$)
   • PR 2: sentiment (~241 tokens ⚠ $\rightarrow \le 200$)
   • PR 3: activity-sweep (~209 tokens ⚠ $\rightarrow \le 200$)

   Phase 2: Body Budget Trimming (Target $\le 5,000$ tokens)

   • PR 4: contributor-to-committer (5,741 tokens ⚠ $\rightarrow \le 5000$)
   • PR 5: nomination (5,610 tokens ⚠ $\rightarrow \le 5000$)

   For every PR, all quoted routing triggers, confirmation gates, byte-identical headings/golden rules, and eval fixtures will be strictly preserved, with measured_tokens stamps updated.
4. Kaap10 commented on Oct 2, 2026

   

   Kaap10

   on Oct 2, 2026

   Contributor

   More actions

   Phase 1 Progress Update (Always-On Frontmatter Trimming)

   Phase 1 is complete — all 3 target skills are now trimmed well below the $\le 200$ token budget with routing triggers fully intact:

   Skill	PR	Before	After	Savings
   onboarding-concierge	#1481	~275 tokens	~133 tokens	-51.7%
   sentiment	#1482	~241 tokens	~131 tokens	-45.6%
   activity-sweep	#1483	~209 tokens	~111 tokens	-46.8%

   Outcome: Saved ~350 always-on tokens, reducing magpie-contributor-growth footprint from ~0.7k to ~0.6k.
5. Kaap10 commented on Oct 2, 2026

   

   Kaap10

   on Oct 2, 2026

   Contributor

   More actions

   Phase 2 Progress Update (Skill Body Budget Optimization)

   Phase 2 is now complete - both over-budget skills in magpie-contributor-growth have been trimmed well below the $\le 5,000$ token ceiling with zero behavioral changes, all companion references intact, and structural surface hashes reconciled:

   Skill	PR	Before	After	Savings	Ceiling Status
   contributor-to-committer	#1487	~5,741 tokens	4,592 tokens	-20.0% (-1,149 tokens)	✅ $\le 5,000$
   nomination	#1489	~5,610 tokens	4,471 tokens	-20.3% (-1,139 tokens)	✅ $\le 5,000$
6. Kaap10 commented on Oct 5, 2026

   

   Kaap10

   on Oct 5, 2026

   Last edited by Kaap10

   Contributor

   More actions

   The optimization recipe for the contributor-growth skill family is fully delivered and merged:

   Phase 1: Always-On Frontmatter Trimming ($\le 200$ tokens)

   • perf(contributor-growth): trim onboarding-concierge routing metadata #1481 — onboarding-concierge (~133 tokens, -51.7%)
   • perf(contributor-growth): trim sentiment skill routing metadata #1482 — sentiment (~131 tokens, -45.6%)
   • perf(contributor-growth): trim activity-sweep skill routing metadata #1483 — activity-sweep (~111 tokens, -46.8%)

   Phase 2: Skill Body Budget Trimming ($\le 5,000$ tokens)

   • perf(contributor-growth): trim contributor-to-committer body budget #1487 — contributor-to-committer (4,592 tokens, -20.0%, brief extraction to companion file)
   • perf(contributor-growth): trim nomination body budget #1489 — nomination (4,802 tokens, -14.4%, companion surface hash reconciled)
7. potiuk commented on Oct 6, 2026

   

   potiuk

   on Oct 6, 2026

   MemberAuthor

   More actions

   Closing: the contributor-growth optimization is complete. Summary of everything that landed:

   Phase 1 — always-on frontmatter (≤ 200 tokens) — thanks @Kaap10

   • perf(contributor-growth): trim onboarding-concierge routing metadata #1481 onboarding-concierge ~275 → ~133
   • perf(contributor-growth): trim sentiment skill routing metadata #1482 sentiment ~241 → ~131
   • perf(contributor-growth): trim activity-sweep skill routing metadata #1483 activity-sweep ~209 → ~111

   Phase 2 — body budget (≤ 5,000 tokens) — thanks @Kaap10

   • perf(contributor-growth): trim contributor-to-committer body budget #1487 contributor-to-committer ~5,741 → 4,592
   • perf(contributor-growth): trim nomination body budget #1489 nomination ~5,610 → 4,471

   Phase 3 — second sweep (#1548)

   • committer-onboarding: split per governance model (asf-pmc / github-codeowners / maintainer-roster) and IP-intake model (icla / dco / no-cla) into detail/, so only the configured model's text is read; routing metadata trimmed; maintainer-approved wording pass. Body 8,042 → 4,722 tokens, 780 → 364 lines (back under the 500-line cap); always-on ~205 → ~159.
   • activity-sweep: four hand-written GraphQL searches (writing to /tmp) replaced by one cached contributor-metrics fetch; counts now dated by the contributor's own activity like the rest of the family.
   • Fixes: private@ self-subscribe wording, gitbox access in the asf-pmc example, and the committer-onboarding step-1 eval (it never asked for JSON, so 6–7 of 9 cases errored on every run; now 26/27 graded).

   Phase 4 — vendor neutrality (#1549)

   • About 30 direct gh calls replaced by contract operations: new read operations on contract:change-request and contract:tracker, repository_metadata / put_file on contract:source-control, and a new contract:people. GitHub is one adapter; Jira is a second tracker backend.
   • contributor-metrics reads code-host and tracker activity separately, so a project reviewing on GitHub with issues in Jira gets correct numbers from both. GitHub-only output is unchanged.
   • Family: capability-pure skills 3 → 6, none vendor-coupled. Evals 145/145.

   Where the family stands now (body tokens, budget 5,000): activity-sweep 3,490 · calibrate 3,730 · candidate-screen 3,695 · committer-onboarding 4,722 · identity-map 3,486 · onboarding-concierge 3,368 · sentiment 4,618 · contributor-to-committer 5,210 · nomination 5,316. Every skill is under the 500-line cap and the always-on budget. contributor-to-committer and nomination have grown slightly past 5,000 again with the later contract and information-only changes; that is small enough to pick up in a future pass rather than keep this issue open.

   Follow-ups, not blocking: GitLab / Forgejo / Bitbucket adapters for the new contract operations, and updating three skill descriptions that still say "GitHub activity" now that the tracker can be Jira.
8. potiuk

   closed this as completedon Oct 6, 2026