docs: refresh roadmap for June 2026

[ralphbean]

Summary

• Restructure Now/Next/Later around outcome-oriented themes derived from 90-day goals and active team work
• Add issue/PR links throughout; flag thin areas with chore issues (chore: build out trustworthiness evidence tracking issues #1991, chore: scope JIRA-driven workflows backlog #1992)
• Drop cross-run memory; add agent attestations to Later

Test plan

• make lint passes
• Team reviews structure after Tuesday planning session

[ralphbean]

This is me trying to get my thoughts together ahead of team planning on Tuesday. We should revise this PR to reflect the maintainers' discussion on Tuesday.

[github-actions]

Site preview

Preview: https://13c088da-site.fullsend-ai.workers.dev

Commit: 78cc1eef26ff0a06e3d7bef0a05a669cbadc6f23

[fullsend-ai-review]

Review

Findings

Low

• [stale-reference] docs/roadmap.md — The diff removes the "Operational observability" section that linked to problems/operational-observability.md. That problem document still exists and is referenced from README.md:37, docs/problems/cross-run-memory.md, docs/problems/adaptive-agent-selection.md, and two ADRs. The practical discoverability impact is minimal since README.md already provides the primary discovery path, but adding a brief mention under an existing section or a one-line Later item would restore navigability.

• [stale-reference] docs/roadmap.md — The diff drops the cross-reference to ADR-0028 from the Forge portability section. ADR-0028 is deprecated, and the new section already links to gitlab-implementation.md, which itself links to ADR-0028 in its opening sentence — so the reference is one hop away rather than missing entirely. Consider adding Related: [ADR-0028](ADRs/0028-gitlab-support.md) alongside the existing gitlab-implementation link for direct navigability.

Info

• [missing-authorization] docs/roadmap.md — No linked issue authorizes this roadmap restructuring. This is a documentation-only PR by a maintainer with the requires-manual-review label, which is appropriate for strategic planning updates.

• [coherence-check] docs/roadmap.md — Cross-run memory removed from Later entirely. The problem doc at docs/problems/cross-run-memory.md still exists and is linked from README.md. The PR body explicitly notes this as intentional.

• [coherence-check] docs/roadmap.md — Operational observability removed from Later. The problem doc remains well-developed with references from multiple ADRs. See stale-reference finding above.

• [technical-accuracy] docs/roadmap.md — The Foundation section drops the full agent enumeration (triage, code, review, fix, retro, scribe) from the opening paragraph. The new "Default agent suite" bullet lists five agents (omitting scribe). The enumeration is available in docs/agents/README.md.

• [scope-update] docs/roadmap.md — Foundation section updates "10+ Konflux repositories" to "20+ Konflux repositories" without citing a source.

• [scope-classification] docs/roadmap.md — The OOTB → default agents terminology change aligns with established usage in docs/agents/README.md and ADR-0003.

• [architectural-alignment] docs/roadmap.md — The restructure from feature-oriented sections to outcome-oriented themes is architecturally sound and aligns with the problem documents and ADRs.

• [scope-consistency] docs/roadmap.md — Removing "Active engagement with additional upstream organizations" from Foundation is coherent — Foundation describes shipped capabilities, not exploratory engagement.

Previous run

Review

Findings

Medium

• [stale-reference] docs/roadmap.md — The diff drops the cross-reference to ADR-0028 that previously appeared in the GitLab/Forge portability section ("The architectural groundwork is laid in ADR-0028"). ADR-0028 exists at docs/ADRs/0028-gitlab-support.md and is referenced from ADR-0036, ADR-0031, docs/plans/agent-execution-environment.md, and docs/problems/gitlab-implementation.md. The roadmap was a discoverable entry point for readers looking for the GitLab architecture decision; dropping it without replacement reduces navigability. Remediation: add Related: [ADR-0028](ADRs/0028-gitlab-support.md) to the "Forge portability" section, alongside the existing link to gitlab-implementation.md.

Low

• [missing-authorization] docs/roadmap.md — No linked issue authorizes this roadmap restructuring. The PR makes strategic priority decisions (restructuring Now/Next/Later, moving BYOA-equivalent work to Now, dropping cross-run memory) without explicit authorization. The PR body references "90-day goals and active team work" but those goals are not linked. This is a documentation-only PR by a maintainer with the requires-manual-review label, which mitigates the concern — consider linking to a planning artifact or documenting in CONTRIBUTING.md that maintainer-driven roadmap refreshes are self-authorizing.

• [internal-consistency] docs/roadmap.md — Several issue/PR references are duplicated across the restructured thematic sections: Forge-portable harness schema (Forge-portable harness schema (role, slug, forge sections with inheritance) #1605, docs: add ADR 0045 for forge-portable harness schema #1848) appears in both "Adoption and extensibility" and "Forge portability"; Schema versioning for harness definition files #235/Config schema validation and versioned migration for .fullsend/ repos #179 appear in both "Adoption and extensibility" and "Versioning and pinning"; Add Renovate automation for openshell version pin #773 appears in both "Agent capabilities" and "Versioning and pinning"; feat(install)!: add --vendor for self-contained workflow and agent assets #1954 appears in both "Adoption and extensibility" and "Quality protections"; feat: default --mint-url to fullsend-ai hosted mint and skip provisioning by default #1145 and feat: default --mint-url to the hosted public mint #2073 are each duplicated between "Agent capabilities" and "Standalone local runtime". While a single issue can serve multiple themes, the repeated references create ambiguity about which theme owns the work. Consider annotating cross-listed items (e.g., "also enables forge portability").

Info

• [list-organization-consistency] docs/roadmap.md — The Foundation "What this phase established" bullet list mixes technical foundations, architectural patterns, and operational achievements without explicit grouping. This is a minor stylistic observation.

• [architectural-alignment] docs/roadmap.md — The restructure introduces outcome-oriented themes that span multiple architectural concerns, departing from the previous feature-oriented organization. This may reduce direct mapping to ADRs but improves clarity for prospective adopters.

• [technical-accuracy] docs/roadmap.md — The K8s section changed "When OpenShell matures" to "When the sandbox runtime matures." Using the architectural abstraction rather than the concrete implementation name is reasonable for a forward-looking roadmap.

• [scope-update] docs/roadmap.md — Foundation section updates "10+ Konflux repositories" to "20+ Konflux repositories" without citing a source.

• [coherence-check] docs/roadmap.md — Cross-run memory removed from Later entirely. The problem doc at docs/problems/cross-run-memory.md still exists and is well-developed. The PR body notes "Drop cross-run memory" but does not explain the rationale.

• [issue-link-density] docs/roadmap.md — Issues chore: build out trustworthiness evidence tracking issues #1991 and chore: scope JIRA-driven workflows backlog #1992 are mentioned in the PR body as chore issues filed for thin areas, but neither appears in the roadmap diff itself.

• [scope-classification] docs/roadmap.md — The change from "OOTB (out-of-the-box) agents" to "default agents" aligns with existing terminology in docs/agents/README.md.

• [terminology-consistency] docs/roadmap.md — Positive: the OOTB→default agents terminology change aligns with docs/agents/README.md usage.

Previous run (2)

Review

Findings

Medium

• [missing-authorization] docs/roadmap.md — No linked issue authorizes this roadmap restructuring. The PR makes strategic priority decisions (moving BYOA and feature refinement to Now, JIRA workflows to Next, dropping cross-run memory from Later, updating deployment scale metrics) without explicit authorization. The PR body references "90-day goals and active team work" but those goals are not linked. Consider linking to a planning artifact or issue that documents the strategic decision, or documenting in CONTRIBUTING.md that maintainer-driven roadmap refreshes are self-authorizing.

Low

• [internal-consistency] docs/roadmap.md — The item "Forge-portable harness schema (Forge-portable harness schema (role, slug, forge sections with inheritance) #1605, docs: add ADR 0045 for forge-portable harness schema #1848)" appears identically in both the "Bring Your Own Agent" section (Now) and the "Forge portability" section (Next). Having the same work item in two time horizons creates ambiguity about when it ships. Consider annotating one reference to clarify the relationship (e.g., "also enables forge portability").

• [internal-consistency] docs/roadmap.md — PR feat(install)!: add --vendor for self-contained workflow and agent assets #1954 is referenced in both "Gradual adoption" (as per-repo installation) and "Quality protections" (as layered/standalone distribution modes). While a PR can serve dual goals, the different framing may confuse readers about the primary purpose. Consider noting the dual relevance explicitly.

• [documentation-consistency] README.md:40 — README.md lists "Cross-Run Memory" as an active problem area with a link to docs/problems/cross-run-memory.md, but this PR removes cross-run memory entirely from the roadmap. This creates an inconsistency where readers may assume all listed problem areas are on the roadmap. Consider adding a note to the README distinguishing problem documents (exploration space) from roadmap items (current priorities).

• [list-formatting-consistency] docs/roadmap.md — The Foundation "What this phase established" bullet list shows improvement by adding technical foundations at the top, but still mixes architectural patterns, operational practices, and deployment metrics without clear grouping. Consider ordering by category: technical foundations first, then operational achievements.

Info

• [technical-accuracy] docs/roadmap.md — The "Kubernetes and OpenShift execution" section changed "When OpenShell matures" to "When the sandbox runtime matures." This generalization is reasonable for a forward-looking roadmap but diverges from ADR-0036 which references OpenShell as the concrete sandbox technology.

• [technical-accuracy] docs/roadmap.md — The Forge portability section (previously "GitLab support") drops the explicit cross-reference to ADR-0028. The ADR still exists and is referenced from other documents. The link was useful navigational context.

• [content-specificity] docs/roadmap.md — The Foundation paragraph removes the agent enumeration ("triage, code, review, fix, retro, and scribe") in favor of the abstract "core agent suite." The new "Default agent suite" bullet partially compensates but lists fewer agents (omits scribe). First-time readers may benefit from seeing the full enumeration at least once.

• [scope-classification] docs/roadmap.md — This change uses the docs: commit prefix, which excludes it from release notes. Strategic priority changes (moving features between Now/Next/Later) may warrant visibility to contributors — consider whether this is intentional for roadmap updates.

Previous run (3)

Review

Findings

Medium

• [missing-authorization] docs/roadmap.md — No linked issue authorizes this roadmap restructuring. The PR makes strategic priority decisions (moving BYOA and feature refinement to Now, JIRA workflows to Next, dropping cross-run memory from Later) without explicit authorization. The PR body references "90-day goals and active team work" but those goals are not linked. Consider linking to a planning artifact or issue that documents the strategic decision, or documenting in CONTRIBUTING.md that maintainer-driven roadmap refreshes are self-authorizing.

Low

• [internal-consistency] docs/roadmap.md — The item "Forge-portable harness schema (Forge-portable harness schema (role, slug, forge sections with inheritance) #1605, docs: add ADR 0045 for forge-portable harness schema #1848)" appears identically in both the "Bring Your Own Agent" section (Now) and the "Forge portability" section (Next). Having the same work item in two time horizons creates ambiguity about when it ships. Consider annotating one reference to clarify the relationship (e.g., "also enables forge portability").

• [internal-consistency] docs/roadmap.md — PR feat(install)!: add --vendor for self-contained workflow and agent assets #1954 is referenced in both "Gradual adoption" (as per-repo installation) and "Quality protections" (as layered/standalone distribution modes). While a PR can serve dual goals, the different framing may confuse readers about the primary purpose. Consider noting the dual relevance explicitly.

• [list-formatting-consistency] docs/roadmap.md — The Foundation "What this phase established" bullet list mixes architectural patterns (sandboxed runner, binary autonomy, trust model), operational practices (dogfooding), and deployment metrics (20+ repos) without consistent grouping. Consider ordering by category: technical foundations first, then operational achievements.

Info

• [technical-accuracy] docs/roadmap.md — The "Kubernetes and OpenShift execution" section changed "When OpenShell matures" to "When the sandbox runtime matures." This generalization is reasonable for a forward-looking roadmap but diverges from ADR-0036 which references OpenShell as the concrete sandbox technology.

• [technical-accuracy] docs/roadmap.md — The Forge portability section (previously "GitLab support") drops the explicit cross-reference to ADR-0028. The ADR still exists and is referenced from other documents. The link was useful navigational context.

• [content-specificity] docs/roadmap.md — The Foundation paragraph removes the agent enumeration ("triage, code, review, fix, retro, and scribe") in favor of the abstract "core agent suite." The new "Default agent suite" bullet partially compensates but lists fewer agents (omits scribe). First-time readers may benefit from seeing the full enumeration at least once.

• [scope-classification] docs/roadmap.md — This change uses the docs: commit prefix, which excludes it from release notes. Strategic priority changes (moving features between Now/Next/Later) may warrant visibility to contributors — consider whether this is intentional for roadmap updates.

[fullsend-ai-review]

🤖 Review · Started 5:06 PM UTC
Commit: ba204cb · View workflow run →

[fullsend-ai-review]

🤖 Review · Started 5:10 PM UTC
Commit: ba204cb · View workflow run →

[fullsend-ai-review]

🤖 Finished Review · ✅ Success · Started 5:16 PM UTC · Completed 5:29 PM UTC
Commit: ba204cb · View workflow run →

[fullsend-ai-review]

🤖 Finished Review · ✅ Success · Started 1:13 PM UTC · Completed 1:27 PM UTC
Commit: 4ed6da4 · View workflow run →

[fullsend-ai-retro]

🤖 Finished Retro · ✅ Success · Started 2:20 PM UTC · Completed 2:24 PM UTC
Commit: 78cc1ee · View workflow run →

[fullsend-ai-retro]

Retro: PR #1993 — docs: refresh roadmap for June 2026

Workflow went well overall. This was a human-authored docs-only PR by ralphbean (4 commits over 4 days, 1 file changed). No triage, code, or fix agents were involved — only the review agent.

Timeline

• June 7: PR opened with initial roadmap draft ahead of team planning.
• June 9: 3 rapid commits after planning session. Two review runs correctly cancelled via superseding; third completed successfully.
• June 10: Final review run completed. Bot approved.
• June 11: Human approved (rh-hemartin). Merged by author.

Observations

1. Review findings went unaddressed. The review agent flagged stale cross-references (README.md still referencing removed sections, dropped ADR-0028 link), duplicate issue numbers across sections, and an unsourced stat change ("10+" → "20+" Konflux repos). None were acted on. The bot approved anyway.

2. Two cancelled review runs from rapid pushes on June 9. The concurrency/superseding logic worked correctly.

No new proposals — existing issues cover all identified improvements

Improvement area	Existing issue(s)
Debounce review dispatch on rapid pushes	#1014, #981
Auto-file tracking issues for unresolved findings	#1956, #1286
Review agent docs feedback quality	#1480
Stale reference scanning improvements	#1061
Use COMMENT verdict (not APPROVE) for human PRs with only medium/low findings	#2115