Stackwich Roadmap #13

This is the working backlog for Stackwich. Start here when choosing the next task.

Research and priorities

This tracking issue is the available fallback for a Projects board: the connected tools cannot create or edit GitHub Projects. Mirror changes here on the Stackwich Roadmap board by hand.

Ready — next release candidates

None.

Ideas — unscheduled

None.

In progress

None.

Parked

• Execute installation fixtures and agent-behavior eval scenarios #12 — Installation fixtures and behavior evals. Deterministic fixtures and CI merged in Check installed scopes against SKILL.md and run deterministic tests in CI #17. The owner decided not to fund the runtime eval harness for now; H2 is parked with it.

Done

• Fix measured reports to count the latest gate and iteration-item state #8 — Fix measured reports to count the latest gate and iteration-item state (PR Count the latest gate and iteration-item state in measured reports #14)
• Document which Stackwich safeguards are instructed versus mechanically enforced #9 — Document which Stackwich safeguards are instructed versus mechanically enforced (PR Document which safeguards are mechanical and which are instructed #15)
• Align release documentation and canonical Stackwich installation links #10 — Release docs, canonical install links, showcase, and published releases v3.1.0–v3.4.0 (PRs Document the canonical install path, version numbers, and the older copy #16, Link the Stackwich showcase from the README #20, Add a release workflow that publishes a release from its CHANGELOG section #28)
• Create a guided first verified task and installation diagnostics #11 — First verified task and diagnostics, run end to end in Claude Code (PRs Add install diagnostics, a safe uninstall, and a first verified task #18, Warn about install permission prompts, from a real walkthrough run #29)
• Bind verification verdicts to the exact revision they tested #19 — Bind verification verdicts to the exact revision they tested (PR Bind verification verdicts to the tree they tested #21)
• Add acceptance-check recipes for browser flows, layout, accessibility, and APIs #22 — Acceptance-check recipes for browser flows, layout, accessibility, and APIs (PR Add acceptance-check recipes for APIs, browser flows, layout, and accessibility #23)
• Work from the user's existing plan, spec, or ticket instead of writing a competing one #24 — Work from the user's existing plan, spec, or ticket; release 3.3.0, policy-rev 6 (PR Work from the user's existing plan, spec, or ticket (policy-rev 6) #25)
• Resume interrupted work from a handoff and check what changed since #26 — Resume interrupted work from a handoff; release 3.4.0, policy-rev 7 (PR Resume interrupted work from a handoff (policy-rev 7) #27)
• .stackwich/ state can be committed into user repositories #34 — .stackwich/ state ignores itself when the tools create it (PR Keep .stackwich/ out of git when the tools create it #44)
• One local command that runs the CI checks #35 — One local command for CI's checks, python3 tools/validate.py (PR Run CI's repository checks with one local command #45; sub-issues Move the skill-lint checks from validate.yml into a script #37, Point CI and CONTRIBUTING at the single check command #38)
• Upgrades leave stale scaffolded agents in place, and nothing detects it #31 — Stale scaffolded agents are detected and refreshed on upgrade (PR Detect and refresh stale scaffolded agents on upgrade #43; sub-issues Detect stale Stackwich agents in check_install and diagnose #32, Refresh existing Stackwich agents in place when upgrading #33)
• policy-rev 8: a leaner, conflict-free policy block #36 — policy-rev 8: one failure ladder, scans to a fork, block cut from 952 to 700 words (PR policy-rev 8: a leaner policy block with one failure ladder (3.5.0) #46; sub-issues Merge the overlapping failure stops into one ladder #39–Release policy-rev 8: ports, revisions, CHANGELOG, README #42)
• Agentic retrieval loop for forked research: rewrite, route, check, retry #30 — Research retries thin results, then reports Sources:, NOT FOUND:, UNCHECKED:; release 3.6.0, policy-rev 9 (PR policy-rev 9: research retries thin results, then names what it didn't check (3.6.0) #47)
• Install docs tell users to run /agents, which Claude Code removed #49 — Install docs no longer point to the removed /agents command; release 3.6.1, no policy change (PR Confirm agents without the removed /agents command (3.6.1) #50)
• Codex installer leaves stale Stackwich agents in place on upgrade #51 — Codex installer refreshes stale Stackwich agents on upgrade; in CHANGELOG Unreleased, no policy or plugin change (PR Refresh stale Stackwich agents on a Codex upgrade #52)

Releasing

Future releases: Actions → release → Run workflow with the version (and the commit to tag for a new version). See CONTRIBUTING.md.

Later — retained research ideas

These are unscheduled ideas, not implementation commitments. Promote an idea to its own issue when it is ready to refine. Research IDs refer to the linked brief.

Most significant

• H1: delivered as Bind verification verdicts to the exact revision they tested #19. Open follow-ups: signed records, and scope checks against a plan.
• H2: reproducible plain-agent versus Stackwich outcome comparisons. Parked with Execute installation fixtures and agent-behavior eval scenarios #12.
• H4: delivered as Add acceptance-check recipes for browser flows, layout, accessibility, and APIs #22. Possible follow-ups: more step types, and an axe-core option.

Medium significance

• M1: configurable and adaptive model selection.
• M2: delivered as Resume interrupted work from a handoff and check what changed since #26.
• M3: delivered as Work from the user's existing plan, spec, or ticket instead of writing a competing one #24.
• M4: automated measurement beyond the reporting fixes already in Fix measured reports to count the latest gate and iteration-item state #8.
• M5: separate solo and team profiles; stronger identity/change-bound approvals when needed.
• M6: additional platforms selected from active user demand.

Low significance

• L1: hosted analytics dashboard.
• L2: expanded agent-role catalog.
• L3: full Jira/Azure DevOps synchronization.
• L4: major visual redesign or additional taxonomy.
• L5: ports without established user demand.

Working rules

1. Capture new ideas using the Stackwich idea/enhancement issue template.
2. Keep significance separate from readiness and release scope.
3. Limit work in progress to one substantial issue.
4. Link implementation PRs to their issues; require the issue's acceptance criteria before closing.
5. Spend 15 minutes weekly reviewing ideas, removing duplicates, and choosing the next task.
6. Update this issue's sections and the individual issue's Workflow status together.

Next action

3.6.1 is published (v3.6.1, policy-rev 9). #51 is merged and listed under CHANGELOG Unreleased; it reaches Codex users when they copy codex/ again, and the next release's notes will include it. Nothing is ready or in progress. #12 stays parked until the owner funds runtime evals. The policy block is at 734 of its 735-word budget (counted with LC_ALL=C wc -w), so the next policy change must cut before it adds. The owner mirrors this issue on the Projects board by hand: the connected tools cannot edit it.