v0.2 — Designs signed off, identities and the session loop in place

Second release of obot.roadmap. Where v0.1 established the hub, v0.2 closes the design phase for the portfolio's founding requirements: designs for the safety.viz library, the histogram pilot, the obot GitHub App, and the new documentation site are all drafted, reviewed, and signed off. Alongside them, the App's two-identity model went live, the working-session routine was codified as reusable skills, and the site picked up an audit-logged roadmap page. Covers 2026-07-03 through 2026-07-09 (40 commits).

Live site: https://jwildfire.github.io/obot.roadmap/

Designs signed off

All four active designs now live under requirements/design/ and publish to the site:

• #1 — safety.viz library and #2 — histogram pilot: ported from the archived obot-claw hub (#11, via #12), test-evidence sections aligned with qcthat's issue-linked test naming (#16), and signed off 2026-07-04.
• #3 — obot GitHub App: drafted (#13), design decisions D1–D4 resolved — the obotclaw app name and a broader permission set (#14) — then updated with as-built registration facts. #3 moved from Design into Development.
• #21 — safety.viz documentation site: drafted and signed off 2026-07-08, specifying the dev/main branch model, three-tier site builds, and gallery screenshots that double as visual-regression baselines.

The obot App identity (#3)

• Two identities now act in the portfolio, codified in AGENTS.md: @jwildfire for interactive work and agent work Jeremy reviews before posting, and obotclaw[bot] (a GitHub App) for automation acting on its own — with local agents minting short-lived App tokens.
• An app-smoke-test workflow verifies the App's token flow end to end (#20).
• Standing operating-autonomy grants are recorded in AGENTS.md: standard roadmap updates need no prior approval, nothing is ever deleted without explicit approval, and review happens on the deployed Pages site.

New requirements

Five requirements joined the roadmap since v0.1:

• #15 — qcthat-compatible QC reporting for safety.viz (JS issue-linked test matrix)
• #17 — rebuild safety.agent as a thin overlay on the gsm.agent harness
• #18 — autonomous obot operations: scheduled and event-driven agent runs
• #21 — safety.viz documentation site: example gallery, test evidence, and API reference
• #22 — R/Pharma 2026 developer-diary blog series

Two new reports feed the decision records behind them: the safety.agent harness proposal (decision record for #17/#18) and the session-orchestration proposal.

The session loop, as skills

The working-session routine is now five repo skills under .github/skills/ instead of ad-hoc habit:

• session-init — open a session with a subagent evidence sweep and a persisted, prioritized kickoff list.
• session-todo / session-update / session-note — render the running list, add tasks, and capture notes mid-session.
• session-wrapup — close out in three phases (collect → discuss → apply), with a three-question checkpoint, ending in that day's diary entry.

The diary convention moved to one post per session (YYYY-MM-DD-N.md for later sessions in a day), and same-day session posts now flow into the News feed. Five new diary entries cover the release window.

Site

• The dashboard page is retired; its key metrics moved to the bottom of the homepage.
• The roadmap page gained a header version badge with an audit-log modal, backed by a semver-bumped roadmap-changelog.json that records every change to what the board shows.
• Site timestamps display in US Eastern Time instead of UTC, and the News type filters got a stylesheet fix.

Full changelog: v0.1...v0.2

---

This release was drafted by Claude Code using Fable 5 and reviewed by @jwildfire