Repository navigation
docs(pi): refresh the pi worked example now the parity gaps have landed - #23152
Conversation
First Principles Review (Opus 5.5) — ⏭️ skippedRevision |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Nothing to check: this is a docs-only refresh, and the claims I spot-checked match main. Pi is in [DESIGN-REVIEWED] e2c57e7 Description read[DESCRIPTION-READ] 05b9fbf635dd3d40642f2a077d97455ab590306d94fc87e56ec3cce11053e14f That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
UX Review (Opus 5.5) — ⏭️ skippedRevision |
Opus 5.5 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
GPT 6.1 Review — ✅ no blocking findingsGPT 6.1 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
iamwhatever
left a comment
There was a problem hiding this comment.
APPROVE. Reviewed head e2c57e7f75893bdfc76b120caaa63a435cd4c4bb.
Docs-only refresh of the pi worked example in harness-onboarding.md. Every membership claim I checked matches main: pi is in ACP_BACKENDS_HONOR_ZERO_TOOL_BAN and ACP_BACKENDS_MEMBER_PANEL, still out of ACP_BACKENDS_MEMBER_CAPABILITIES and ACP_BACKENDS_MEMBER_TOOL_LIMITS; test/real_adapters/package.json pins pi-acp 0.0.34 and pi 1.3.0 and the Real Adapter lane drives test_pi_real_adapter.py; pi_node.py and the seven live captures exist. The gaps left open are tracked by open PRs #22911 and #22923. AI lanes pass and all checks on this head are green.
— crew-pr-reviewer (r121)
Problem / Motivation
Goal: The pi worked example in
harness-onboarding.mddescribes what main's pi support actually does today, instead of listing as open three gaps that have since landed.The "Worked example: the Pi harness" table's
open gapsrow still says pi is out ofACP_BACKENDS_HONOR_ZERO_TOOL_BANandACP_BACKENDS_MEMBER_PANEL, and that CI installs neitherpinorpi-acp. On main all three are no longer true: #22808 added pi to the zero-tool-ban set, #22813 added it to the member-panel set, and #22807 pinnedpi-acp+ pi intest/real_adapters/and drives them in the Real Adapter Contract Tests lane. The3 spawn pathand8 live spillrows also predate the node-engine guard (#22883) and the two new live captures.Why it matters
This section is the reference a reader uses to onboard the next harness and to see where pi stands. A row that calls merged work "open" sends readers chasing gaps that are closed and hides which ones are still open.
Not a goal
ACP_BACKENDS_MEMBER_CAPABILITIES,ACP_BACKENDS_MEMBER_TOOL_LIMITSand all three steer sets. Those sentences stay in the row word for word, so each of those PRs can update its own sentence when it lands.harness-parity.mdor any other doc.What changed (motivation → approach → change)
Stale rows → checked each claim against main's set definitions (
agent_sdk/backends_mcp.py,backends_members.py,backends_turns.py),test/real_adapters/package.json,.github/workflows/ci.ymlandtest/fixtures/acp_frames/pi/→ moved each landed gap into the row that owns it, in the section's existing style:2 capability sets: adds pi's membership inACP_BACKENDS_MEMBER_PANEL(own evidence: the handoff carries the panel element,member-panel-mount-live.jsonl) and inACP_BACKENDS_HONOR_ZERO_TOOL_BANas its one mirror-less member (_mirrorless_member_mountsreads"tools": [],_deny_zero_toolsrefuses each gate dialog,zero-tools-refused-live.jsonl).3 spawn path: adds the Nodeengines.nodeguard (acp/harness/pi_node.py, fix(pi): run pi under a Node that meets its engine floor #22883): the fallback order, the refusal, the confined probe, and which cases it leaves to the read-back.8 live spill: seven live captures instead of five; adds the Real Adapter lane, which pinspi-acp0.0.34 / pi 1.3.0 and drivestest_pi_real_adapter.pythroughPiLaunch.resolve_spawnon every backend PR.open gaps: drops the zero-tool-ban, real-adapter-lane and member-panel entries. Saved-spec verification, the spec's own MCP servers / per-tool deny set (projection stillbroker-only;ACP_BACKENDS_MEMBER_TOOL_LIMITSis still{kiro}), and mid-turn steer stay in the row, unchanged.Not every gap is closed, so the row stays rather than being marked "at parity".
Heads-up for #22911: it edits the same
open gapsline, so its next rebase will hit a one-line conflict here. Resolving it means keeping this PR's row and swapping in #22911's saved-spec sentence. #22923 does not touch this row.Backwards compatibility
Compatible: documentation only; no code, config or test changed.
Removes nothing: the only text removed is three
open gapsentries describing state main no longer has; each is restated as current state in the row that owns it.Tests
N/A — docs only. Run locally:
./scripts/docs-lint.sh(all documentation checks passed), andpytest test/test_backend_id_uniqueness.py test/test_provider_mirrors.py test/test_harness_parity.py(the tests that read this doc or pin these sets; 153 passed).Manual verification
Checked every membership claim in the edited rows against main at 69b2e50: pi is in
ACP_BACKENDS_HONOR_ZERO_TOOL_BAN,ACP_BACKENDS_MEMBER_PANELandACP_BACKENDS_MEMBER_DISPATCH; it is not inACP_BACKENDS_MEMBER_CAPABILITIES,ACP_BACKENDS_MEMBER_TOOL_LIMITS({kiro}),ACP_BACKENDS_STEER,ACP_BACKENDS_STEERING_REQUESTorACP_BACKENDS_STEER_CONCURRENT_PROMPT; its projection isProjectionKind.BROKER_ONLY. Sensitive Change Review: the one changed path matches none ofSENSITIVE_GLOBS.Related Issues
Follows #22807, #22808, #22813, #22883, #22951.
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)Contribution License Agreement
Contributions are accepted under the project's Apache-2.0 license; see the Licensing section of CONTRIBUTING.md.