OpenSiro's public record of improvements to other open-source projects. We trace a contribution from a reproducible problem and source-native fix through its external pull request, maintainer feedback and independently verified outcome. A prepared patch is not a merged contribution, and a merged PR is not automatically a demonstrated product benefit.
Contribution protocol: SemVer 0.11.0 · numbered changes · versioning policy. Versions track the contribution procedure, not Profile/Skills methodology or individual PR outcomes.
| I want to… | Go to |
|---|---|
| Understand the broader OpenSiro project | Visual VSM map |
| See contributions and what happened to them | Unified PR + verified-outcome report (latest successful observe summary/artifact) · Public portfolio · Verified merged changes · Historical dashboard · Harness status |
| Inspect actual third-party PRs | Citadel #366 · Agent Fleet #43 · PR observation and verification rules |
| Review source-backed evidence | Casebook · Native test pilots |
| Contribute or maintain this project | CONTRIBUTING.md · UPSTREAMING.md |
The dashboard is a frozen eight-case historical projection; the portfolio and daily unified report are projections, not authority for a third-party PR's live state. Consult the exact GitHub PR for current disposition. Our verified-outcomes process distinguishes machine evidence candidates from accepted outcomes; the Organization collector contract now accepts only live-verified merged-change evidence. It does not claim deployed benefit or productivity; missing observations remain unclaimed.
One route: OpenSiro START_HERE → ecosystem ownership → AGENTS.md → UPSTREAMING.md → exact current issue/PR. Use the Organization TODO only for selected shared work, not as a substitute for the owning issue. New or cross-domain tasks start with CONTRIBUTOR_START.
For the common operations:
| Work | Procedure |
|---|---|
| Investigate a new useful native capability | Capability-first research · Assessment candidate batches |
| Prepare source code, native tests, and the fork-local PR | Early working-fork check · Working-fork pipeline |
| Publish or comment to a third-party project | Current external-public communication consent gate · revised standing publication policy |
| Observe feedback, states, and outcomes | PR maintenance · Verified outcomes |
| Operate this repository's own Actions | Actions map · Operations |
Outbound consent: a public comment, issue, PR, review, or design question in a third-party repository always requires prior explicit owner confirmation for the exact post. This rule supersedes the older standing no-prompt behavior. Internal OpenSiro work remains governed by its own permissions.
Where work belongs: actual patches, tests, and proposed external PR bodies belong in the matching opensiro-contributions/<target> fork. This repository holds governance, historical test evidence, registries, observations and derived dashboards. Source-of-truth contract. Index owns canonical VSM assessments; Skills owns their methodology; Profile owns VSM semantics.
An Index assessment can lead to the strict assessment-derived S1 test; the separate generic OSS preview does not prove an assessment-derived fix. The historical OSM delegation design does not grant host execution rights; see automatic fork executor recovery for the specific technical gate.
Effective standing preparation permits bounded source/fork preparation. The older standing publication mandate is narrowed by the owner's 2026-11-15 instruction: before every public communication or PR publication to an outside project, first show the exact target and message/PR diff and obtain current, explicit per-action owner approval. External-public communication gate is fail-closed; continue, generic standing authorization, previous approvals, and passing tests never authorize the outward write. Credentials, source-native CI, legal and third-party authority requirements remain separate. Executor gate #6.
For new upstream upgrade batch research under protocol v0.11.0 (retaining the v0.9.0 capability-first research stage), first specify a new operator task, output and bounded benefit (capability-first contract), keeping reliability repairs distinct. The earlier v0.8.0 six-function depth requirement remains: a six-function source screen is not a concluded product-opportunity investigation: use discriminating post-source evidence and resolve the strongest need=PENDING / counterfactual=PENDING native falsifier when feasible, or preserve an exact tested/blocker follow-up. Distinguish SCREENING_COMPLETE from OPPORTUNITY_RESEARCH_PENDING in prose; the four native-fit admission gates and all external rights remain unchanged. See batch 39 diagnosis and procedure.
For upstream upgrade batch, every selected repository needs an S1/S2/S3/S3*/S4/S5 scan and six concrete post-source scoped probes, not merely a filled preregistration matrix. The command can prioritize source-owned maintainer/issue leads over latest Index review (disclosed as source-first / not strict-derived), or preregister an independent Index-first hypothesis before fresh source inspection. Exploratory need remains pending until tested; four native-fit PASS gates and independent validation still precede any accepted improvement claim. New capabilities have priority, supported repairs remain useful; an explicit batch of ten means ten repositories × six evidence-backed function dispositions, not a patch quota.
python3 -m unittest discover -s tests -v
python3 scripts/upstream_status.py --check-generated
python3 scripts/harness_status.py --check
python3 scripts/portfolio_reconcile.py --validate-only
python3 scripts/verified_external_outcomes.py --validate-onlyThe unfiltered repository-integrity workflow runs unit tests and generated-inventory checks on every PR. It is the required GitHub Actions check on protected main under Organization #305; an unrelated failing PR is blocked, not bypassed. Third-party source-native CI belongs in working forks; migration #214 has completed placement, not universal native parity.
Legal boundary: no repository-wide root LICENSE has been adopted. Standards #130 and license draft PR #1 require the actual rightsholders' decision; do not relicense third-party code or original prose by inference.