Repository navigation
docs(rfc): propose chat_rename for self-rename - #33633
Conversation
34fcea5 to
9974107
Compare
|
Review input on the RFC's two open questions. This is input only, not approval of the design. I checked the "Current state" claims against
Open question 1: should Open question 2: conductors. Keeping the tool visible behind a prompt costs no new code. It is also refused on the turns a conductor mostly runs, since a patrol or wake is not a person's turn. A per-tool deny would add a mechanism the conductor specs don't have today, for a case that already needs a person's turn and an approval click. So my input leans toward keeping it. However, A small note on Migration: the exit criteria name no test for the allowlist case, where a titled session has no recorded origin and the slot is not yet reloaded. That case is the reason A design is agreed to by a human, not by an agent. A Kiro Crew team member needs to sign off on this design and on open question 2 before the implementation in #33736 moves forward. AI-assisted, sent by CommunityCrew |
First Principles Review (Opus 5.5, fork) — ⏭️ skippedRevision |
UX Review (Opus 5.5, fork) — ⏭️ skippedRevision |
Design Review (Opus 5.5, fork) — ✅ PASSDesign-level review of Design-Verdict: PASS Docs-only RFC. The design is proportionate and built on existing gates, and nothing is taken away. Notes
[DESIGN-REVIEWED] 500abac Description read[DESCRIPTION-READ] 5a7e70dc18669920bea52fee073ddfab22da7c2b3337b44230249a0683516359 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 6.1 Review (fork) — ✅ no blocking findingsReviewed Review detailsNo findings. |
Opus 5.5 Review (fork) — ✅ no blocking findingsReviewed |
9974107 to
1fdfcc5
Compare
|
Thanks @bolichen97. I revised the RFC for all three points:
The same push covers the Design Review's two notes: the Goals now say a rename happens only on a person's turn, and the @iamwhatever, since the conductor entry is yours, could you or another maintainer sign off on the design and on open question 2 when you have a moment? I'll record both in the RFC and move it to AI-assisted, sent by Kiro Crew |
|
Thanks for the revision. The branch now conflicts with The Sensitive Change Review and Fork PR Description failures on this head came from a GitHub API rate limit, not from your change. The design still needs a Kiro Crew team member's sign-off, as the new Sign-off section says. AI-assisted, sent by CommunityCrew |
1fdfcc5 to
bea8701
Compare
|
Thanks @bolichen97. Done: rebased onto current AI-assisted, sent by Kiro Crew |
|
This RFC needs a design sign-off from a maintainer before it can merge. A person agrees to a design, not an automated review. Two decisions stay with maintainers. The first is the Review result on head Done well: the origins an agent may replace are an allowlist. A title that has no recorded origin therefore stays protected, even before a reload. For the security review:
Optional:
|
bolichen97
left a comment
There was a problem hiding this comment.
Read the whole diff at bea8701: the RFC and its README row (the long README paragraph differs only by the new row's sentence). This is a design review comment, not an approval. The sign-off on the tool, and the separate one on conductors, belong to the maintainers.
Design brief
What it is: a new chat_rename(title) directive on the core server, shaped like chat_tag. It renames only the session whose turn produced it, and only on a turn a person sent (it joins USER_SURFACE_DIRECTIVES). It goes through session_rename's title rules and apply_manual_title. A title a person set (origin user, or one with no recorded origin) is never replaced. The agent's own title is stored with a new origin, agent, which the auto-titler leaves alone and the person's next rename overwrites.
Pros:
- Fixes a real gap: on the default spec an agent cannot follow a person's naming convention, because
session_renamelives in the opt-in dashboard server and needs atargetthe agent cannot learn. - No new reach. There is no target argument, and it applies to the slot the turn already holds, which also removes the key-reuse window seen in #20064.
- It cannot replace a name a person chose, and the allowlist of replaceable origins protects writers that record no origin.
- Additive: no existing tool, route or default changes, and an older version reads
agentasuser, so a downgrade stays safe.
Cons:
- One more tool schema on the surface every session carries, and one more
title_originvalue that readers must accept. - A title an agent set through
session_renameis written asuser, sochat_renamecannot replace it. The RFC states this cost (open question 1). - The tool reply only says a rename was requested, because the write happens after the tool returns. The agent has to report a request, not a finished rename.
- Injected content inside a person's own turn can still steer a rename. The RFC bounds this to a label on the person's own sidebar that they can change.
Long-term debt: low. It reuses the title rules and the write path. The one lasting cost is the agent origin, which every future title writer has to place in the origin table.
My stance: support item 1, the tool as designed. Item 2, conductors reaching it behind an approval prompt as they reach chat_tag today, is an owner decision under the 2026-11-04 conductor rule. The proposed answer (keep it, no new code, patrol and wake turns are refused anyway) looks right to me.
Small points on the text, none blocking:
- The Summary says "Not behind
agent.session_control". Please say once in the Design section that the tool needs no setting to turn it off, so a reader sees the rollback is reverting the PR.
|
Thanks @bolichen97 for the design review, and for the review notes above. Each is fixed in the RFC text, committed locally:
To keep this head green while it is under review, I'll push these together with the sign-off record rather than reset CI and the run approvals now. If you'd rather read them first, say so and I'll push right away. AI-assisted, sent by Kiro Crew |
bea8701 to
84ed02e
Compare
|
Pushed #34138 (merged 03:37 UTC) changed the write this RFC builds on.
The same push carries the three text fixes from my comment above, plus the crewmate-thread exception in Summary item 4. Every code reference was re-checked against @bolichen97, the new head's fork runs need approval again. @iamwhatever, the two sign-off items are unchanged. AI-assisted, sent by Kiro Crew |
|
Superseded: @bolichen97, thanks for approving the runs. CI on Could a maintainer re-run these? No new push is needed.
AI-assisted, sent by Kiro Crew |
An agent on the default spec cannot rename the session it runs in: session_rename lives in the opt-in kirocrew-dashboard server, which the default spec does not mount, and it needs a target the agent cannot learn. This RFC proposes chat_rename(title), a kirocrew-core session directive shaped like chat_tag that renames only the calling session, on a turn a person sent, through session_rename's own title rules and refusals.
84ed02e to
500abac
Compare
|
Pushed
@iamwhatever, the two sign-off items are unchanged: the tool, and conductors reaching it only through an approval prompt. @bolichen97, could you approve the fork runs for the new head when you have a moment? AI-assisted, sent by Kiro Crew |
bolichen97
left a comment
There was a problem hiding this comment.
Approving at 500abac. I read the whole diff: the RFC and its README row.
Maintainer decision on Sign-off item 1 (the tool): accepted as designed. chat_rename(title) is a core directive shaped like chat_tag, for every spec that mounts @kirocrew-core.
- It renames only the calling session, with no target, and only on a person's turn.
- That turn needs all three of main's own checks: a person's actor,
slot._turn_user_sent, and no channel provenance. A cron turn is refused, including a script cron'ssend_to_sessionturn, and so are sub-agent, wake and channel turns, and a person's turn with a peer's, an app's or a channel's text steered into it. - It goes through
session_rename's title rules andapply_manual_title, so it writes the samechat.manual_renameaudit row. - It never replaces a title a person set:
_AGENT_REPLACEABLE_ORIGINSis an allowlist, so a writer that records no origin is protected too. - It refuses a
channel_origintab, bound or not. - It stores the new origin
agent, which the auto-titler leaves alone and which is never read as Slack mrkdwn.
It adds a capability the default agent lacks today, and it reaches no other session, so nothing main allows is refused. Open question 1 (leave session_rename writing user) is fine as proposed.
Not decided here: Sign-off item 2, whether a conductor may reach chat_rename behind its approval prompt. Under the 2026-11-04 conductor decision that is a separate owner decision, so this approval does not cover it.
Doc only, and all 94 checks on this head are green.
Bolin Chen accepted the chat_rename tool as designed in an approving review of kirodotdev#33633. Record it in the RFC's Sign-off section and its index row, and move the status to in-progress, since kirodotdev#33736 implements it. Whether a conductor may reach the tool behind its approval prompt stays open for the owner: the review leaves it out.
Bolin Chen accepted the chat_rename tool as designed in an approving review of kirodotdev#33633. Record it in the RFC's Sign-off section and its index row, and move the status to in-progress, since kirodotdev#33736 implements it. Whether a conductor may reach the tool behind its approval prompt stays open for the owner: the review leaves it out.
Bolin Chen accepted the chat_rename tool as designed in an approving review of #33633. Record it in the RFC's Sign-off section and its index row, and move the status to in-progress, since #33736 implements it. Whether a conductor may reach the tool behind its approval prompt stays open for the owner: the review leaves it out.
Problem / Motivation
Goal: Record a design for letting an agent rename the session it is running in on the default agent, without ever replacing a name a person chose, so the implementation can be reviewed against an agreed decision.
An agent on the default spec cannot rename its own session. The only agent-facing rename,
session_rename, lives in the opt-inkirocrew-dashboardserver, which the default spec does not mount, and it takes atargetthe agent has no way to learn. Fixing that means adding a tool tokirocrew-core, the surface every session carries, and a value to the storedtitle_origin. Giving the default agent a new ability is a product-shape decision, which CONTRIBUTING.md lists as a large change, so the design comes first, in its own pull request.Why it matters
People find their work by the sidebar title, and many keep a naming convention in their own instructions (#13797's example is
JIRA-XXXX — <ticket title>). Generated titles are often weak, which is why the dashboard has "Regenerate title"; the agent working in a session can name it better, but only if it never overwrites a name the person typed. The implementation also needs this decision onmain: the First Principles lane reads an RFC's status from the base commit.Not a goal
What changed (motivation → approach → change)
This adds
docs/request-for-change/rfc-agent-self-rename.mdwith statusdraft, and its row in thedocs/request-for-change/README.mdindex.The RFC proposes
chat_rename(title), akirocrew-coresession directive built likechat_tag. It renames only the session whose turn called it, only on a turn a person sent, throughsession_rename's title rules, refusal codes and write.It records three decisions with their reasons. The agent never replaces a title someone set: it may replace only a generated title, a crewmate thread's name or its own earlier rename, which records a new origin,
agent; a person's rename replaces it and "Regenerate title" hands the name back. Incognito and temporary sessions can rename themselves, since their refusal exists so other sessions cannot address them. And the name followschat_tag, the core's other self-only directive, rather than the dashboard server'ssession_*verbs. It also records the rejected alternatives: mountingsession_renameon the default spec, asession_rename_selfin the dashboard server, atarget-lesssession_rename, and a naming pattern in the automatic titler.It asks for two sign-offs, recorded separately: the tool itself, and, as the explicit owner decision the 2026-11-04 conductor decision requires, whether a conductor may reach it behind its approval prompt. Two open questions carry proposed answers:
session_renamekeeps writinguser, whose cost the RFC records under Consequences, and conductors keep the tool behind the prompt rather than gaining a per-tool deny. Once both are signed off, this pull request records them in the RFC and its index row and sets the status toin-progress, since the implementation is already open as #33736; that implementation's First Principles lane then reads the decision frommain.Backwards compatibility
Compatible: documentation only. No code, schema or default changes.
Removes nothing: the diff adds one document and one index row, and appends one sentence to the index's audit note.
Tests
N/A — documentation only.
scripts/docs-lint.shpasses on this head.Manual verification
N/A — documentation only; every code reference in the RFC was re-checked against
mainat77fabdd1f.Related Issues
Part of #33624
Implementation: #33736. Related: #13797, #15164, #17585, #20064, #27705, #13144.
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)