You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
After Part 3 (skills infrastructure). testing builds on the harness design of #4854 and #4855.
Scope
Adapt from the Inspector (§6 of the doc):
project-structure: each server's layout, and where a new file goes.
local-dev: npm workspaces and uv; running each server over the transports it implements (stdio for all seven; everything also serves SSE and Streamable HTTP); local npx/uvx builds; pin with overrides, never npm audit fix.
testing: in-process protocol harnesses (an SDK Client over an in-memory transport for TS; ClientSession for Python), test placement, the coverage gate.
client-smoke: drive a server with the Inspector V2 (web and CLI) and an LLM client, in both spec eras (2026-07-28 and 2025-11-25, Tracker: 2026-07-28 Spec Refactor #4857). This replaces the Inspector's test-servers skill. Driving the Inspector's web client needs browser automation, so also enable the Playwright plugin in .claude/settings.json, or document the Inspector CLI as the scripted path and the web client as the hand-driven one.
Acceptance criteria
All four skills are merged with eval cases.
A full skills:eval re-run shows no regression in the skills that already exist.
If this is the first skill-adding PR to merge, remove Part 3's empty-skills bootstrap allowance from verify:skills.
Branch
Targets v2/main. Closes #N won't auto-close an issue on a non-default branch, so close this one by hand on merge and move the card to Done.
Part of the agentic software factory tracker #4858, Wave 4. Proposed as S11 in
docs/agent-guidance-inception.md(PR #4861) §9, from #4859.Depends on
After Part 3 (skills infrastructure).
testingbuilds on the harness design of #4854 and #4855.Scope
Adapt from the Inspector (§6 of the doc):
project-structure: each server's layout, and where a new file goes.local-dev: npm workspaces anduv; running each server over the transports it implements (stdio for all seven;everythingalso serves SSE and Streamable HTTP); localnpx/uvxbuilds; pin withoverrides, nevernpm audit fix.testing: in-process protocol harnesses (an SDKClientover an in-memory transport for TS;ClientSessionfor Python), test placement, the coverage gate.client-smoke: drive a server with the Inspector V2 (web and CLI) and an LLM client, in both spec eras (2026-07-28 and 2025-11-25, Tracker: 2026-07-28 Spec Refactor #4857). This replaces the Inspector'stest-serversskill. Driving the Inspector's web client needs browser automation, so also enable the Playwright plugin in.claude/settings.json, or document the Inspector CLI as the scripted path and the web client as the hand-driven one.Acceptance criteria
skills:evalre-run shows no regression in the skills that already exist.verify:skills.Branch
Targets
v2/main.Closes #Nwon't auto-close an issue on a non-default branch, so close this one by hand on merge and move the card to Done.