Repository navigation
docs(rfc): automatic stale-config session reload at the turn boundary - #15409
billygerhard wants to merge 1 commit into
Conversation
Design Review (Fable 5, fork) — 🟡 CONCERNSDesign-level review of Design-Verdict: CONCERNS One exit criterion commits to withholding a user's message via a "post-spawn re-read" mechanism the Design section never defines and §2a appears to forbid. WatchThe migration-plan criterion "A turn not run because a post-spawn re-read found an unapproved edit writes a notice that names the message as not processed" is the only place the document allows a turn to not run; §2a states "The check never fails a turn on its own account," and the pre-teardown/post-spawn re-reads appear only in a passing Security-considerations sentence ("stay as defence in depth for that press") with no definition of when they run, what an "unapproved edit" is, or why dropping the message is the right outcome. Merging this PR is, per the description, maintainer acceptance — so #14810 would trace a message-dropping behavior to a decision that was never actually specified. [DESIGN-REVIEWED] 59ada50 Description read[DESCRIPTION-READ] 88fed535558096d27dedb7c2606a20704ec8519257407e22af37762b4d12ffc1 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 |
GPT 5.6 Review (fork) — ✅ no blocking findingsReviewed Review detailsNo findings. |
dd7d3b7 to
8e304b1
Compare
Opus 5 Review (fork) — ✅ no blocking findingsReviewed Review detailsTwo added doc claims contradict the code: the RFC's security analysis says the OS sandbox does not deny paths FINDING — docs/request-for-change/rfc-stale-config-session-reload.md:355 — "the OS sandbox does not deny these paths" (and at :364, crew FINDING — docs/request-for-change/rfc-stale-config-session-reload.md:282 — the exit criterion "A turn not run because a post-spawn re-read found an unapproved edit writes a notice that names the message as not processed" names a mechanism the Design never defines ( [OPUS-REVIEWED] 59ada50 |
8e304b1 to
269b82b
Compare
269b82b to
080b072
Compare
080b072 to
8736fa0
Compare
8736fa0 to
4a4c098
Compare
4a4c098 to
7fc9cbd
Compare
7fc9cbd to
c28356b
Compare
c28356b to
0ea64f7
Compare
0ea64f7 to
df33241
Compare
42a3b3f to
2eef70c
Compare
2eef70c to
cbd3682
Compare
cbd3682 to
e282184
Compare
e282184 to
0614147
Compare
0614147 to
d7f279f
Compare
Record the product-shape decision behind kirodotdev#14810: a chat whose spawn-time config changed reloads its agent process at the start of its next turn, keeping the conversation, only when idle with no sub-agents attached; otherwise the reload waits and the Reload row shows it. Refs kirodotdev#10211.
d7f279f to
59ada50
Compare
|
Closing: the implementation PR #14810 changed scope. It no longer reloads chats automatically; it only shows a stale-config badge and leaves reloading to a manual press, as today. That is a small UI addition with no product-shape change, so this RFC for the automatic-reload design is no longer needed. |
Problem / Motivation
Goal: Record the design decision that a chat automatically reloads its agent process onto changed config at its next turn, so the implementation in #14810 traces to a decision on main.
A chat's agent process reads its MCP servers, agent spec,
mcp.json, ACP backend and MCP gateway stubs once, when it starts. Edits made after that never reach it, and nothing tells the user. The only remedy today is the manual Reload action. #14810 makes that reload automatic at the next turn boundary, which changes a default for every chat, so perdocs/request-for-change/README.mdthe decision has to be on main first.Why it matters
Without a recorded decision the First Principles lane blocks #14810, which closes #10211. The document also gives maintainers one place to accept or reshape the default (for example, ask for an opt-out setting) before the code lands.
Not a goal
What changed (motivation → approach → change)
Adds
docs/request-for-change/rfc-stale-config-session-reload.md(Summary, Motivation, Goals, Non-goals, Design, Migration plan with exit criteria, Backward compatibility, Security considerations, Alternatives considered, Open questions) and its index row plus audit sentence indocs/request-for-change/README.md.Design in one paragraph: fingerprint a chat's spawn-time config inputs; at the start of its next turn, if the fingerprint is stale and the chat is idle with no sub-agents attached, reload through the Reload action's own path (conversation preserved) and show the existing
session_reloadnotice; otherwise send the turn on the current process, mark the reload pending, and show it on the Reload menu row. Edits kiro-cli 2.21+ already hot-reloads do not trigger a reload. Asession_reloadMCP tool exposes the same state and lets an agent request a reload (performed at its next turn boundary) through the session-control API and its principal gates.Status is
in-progressbecause the implementation is live in open PR #14810, matching how other rows in the index record a decision whose implementation is in flight. Merging this PR is the maintainer acceptance of that decision; if maintainers want a different shape, I'll revise the document and #14810 together.Backwards compatibility
Compatible: documentation only.
Tests
N/A — documentation only.
python3 scripts/docs_lint.pypasses.Manual verification
Every symbol the status line and index row cite was checked on
aec81410e2:api_chat_slot_reloadinsrc/kiro_crew/dashboard/chat_handlers.py,SESSION_RELOAD_KINDinsrc/kiro_crew/dashboard/system_notices.py,MCP_HOT_RELOAD_MIN_KIRO_CLI_VERSIONinsrc/kiro_crew/mcp_hot_reload.py,_reset_all_sessionsinsrc/kiro_crew/dashboard/handlers/sessions.pyand its three callers (handlers/agents.py,handlers/mcp.py,handlers/computer_use.py),session/set_modelinapi_chat_slot_model, and nostale_configmodule undersrc/kiro_crew/dashboard/.Related Issues
Refs #10211. Implementation: #14810.
Checklist