Release 0.3.0 — Channel projection + chat-of-record + merge-hand contract

[daxxsyntax]

Summary

Three epics shipped on release/0.3.0:

• Chat-of-record (PR Chat-of-record: mandatory posting + #atlas history + sudden-death recovery reader [sp-12f09190] #54, ryve-12f09190): ryve post + ryve channel tail CLI primitives, #atlas well-known workshop channel, ryve assign close gate refusing zero-post closures, boot-prepend of recent channel context in hand_spawn, MCP chat.post + chat.tail wrappers, Atlas/Head/Hand prompt updates referencing the mandatory posting boundaries.
• Channel projection (PR crew cr-43e1e251: channel projection — query module + UI panel [sp-06816a07] #53, ryve-06816a07): 5-axis query module (epic/spark/assignment/PR/actor) with FTS5 search + mention override, filter preset persistence with unread tracking, live UI panel with preset picker + search + unread badges.
• Merge-Hand contract (PR crew cr-f9bb4626 v2: 0.3.0 merge-hand-contract — manual integration with conflict resolve #55, ryve-476ef264): new merge_hand role with prompt composer, precondition checker (all Approved + no conflicts + CI green + zero Stuck), Sub-PR → Epic merge flow in Assignment-creation order with git merge --no-ff, conflict handoff via transition validator (Approved → Rejected reason=conflict), Epic PR lifecycle driving Epic → main merge with reviewDecision + statusCheckRollup gates and mark_assignment_merged pinned to merge_hand role, E2E test covering the full happy path.

Merge commits

• cca8420 chat-of-record (PR Chat-of-record: mandatory posting + #atlas history + sudden-death recovery reader [sp-12f09190] #54)
• f07c870 channel projection (PR crew cr-43e1e251: channel projection — query module + UI panel [sp-06816a07] #53)
• 0a24ae2 merge-hand contract (PR crew cr-f9bb4626 v2: 0.3.0 merge-hand-contract — manual integration with conflict resolve #55)

Copilot review addressed across the three PRs

24 comments total (7 on #54, 13 on #53, 4 on #55) covering real correctness bugs (parent-epic fallback in count_posts_since_claim, FTS validation in preset_unread_count, preset-filter clear semantics, default-channel status filter, mention-override gated on explicit nick, sender-label prefers metadata.actor_id, SelectTab IRC refresh) and known-incomplete items tracked as follow-up sparks (N+1 in unread counts + precondition checker, JumpToTail scroll wiring, Task-chain for initial unread badge load).

Known tech debt from this release

1. Dead-Head-after-dispatch pattern. All three 0.3.0 Heads exited silently after dispatching their wave-1 / wave-2 children (claiming a cron poll that was never created). Manual intervention pushed Hand branches + spawned Mergers. Atlas archetype should either (a) genuinely set up a cron, or (b) block on child closures instead of pretending it set up a watcher.

2. Base-branch stacking failed on merge-hand-contract's Epic-PR child. sp-9f6b0c96 (Epic PR lifecycle) branched off main without stacking onto sp-6b261ad0 (Sub-PR merge flow), producing an unresolvable-by-automation merge conflict. The blocks bond was declared but the Hand spawn resolver didn't find a branch tip to cut from at the moment of spawn. Needs investigation — likely a timing/race in resolve_hand_base_ref.

3. Merger defaults to main for the base branch (recurring). Mergers on #53, #54, #55 all opened against main by default instead of release/0.3.0. Retargeted by hand each time. Same issue as 0.2.1; still unfixed. The Merger prompt needs to learn about the release branch from the release row or parent epic's release bond.

4. Format check uses stable rustfmt but .rustfmt.toml has unstable features. Local nightly cargo fmt produces a different output than CI's stable cargo fmt --check. Surfaced on PR crew cr-f9bb4626 v2: 0.3.0 merge-hand-contract — manual integration with conflict resolve #55 and fixed in 058d5e9 with another fmt pass.

Test plan

• ryve init a fresh workshop. ryve post --channel '#atlas' 'hello' then ryve channel tail --channel '#atlas' --limit 5 — the post round-trips through irc_messages.
• Claim a spark under an epic, then try ryve assign close <session> <spark> without posting. Refuse with a clear message. Post once, retry — closes.
• Spawn a Hand on an epic child. Kill its session mid-task, re-spawn. The new Hand's prompt contains the prior session's posts as context.
• Open the workshop UI. Click the IRC view tab. See the projection panel with filter chips, preset picker, unread badges. Create a preset, switch to another channel and back — preset state persists.
• Run ryve head spawn <epic> --archetype build --agent claude on an epic with at least 2 blocks-bonded children that have both landed. The MergeHand contract's precondition checker admits the spawn when all children are Approved.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

Release 0.3.0 integration delivering chat-of-record (durable IRC-backed posting + tailing + close gate), channel projection (filtered IRC views + presets + unread), and the MergeHand contract (new role + locked transitions + E2E coverage) across the ipc, data, UI, and CLI layers.

Changes:

• Add chat-of-record primitives (ipc::chat_of_record), CLI commands (ryve post, ryve channel tail, ryve assign close gate), and MCP wrappers (chat.post, chat.tail).
• Introduce channel projection query engine + persistence (presets/unread) and a new UI IRC projection tab with periodic refresh.
• Implement MergeHand transition constraints (conflict handoff + merge-only “Merged” transition) and add end-to-end tests for the happy path.

Reviewed changes

Copilot reviewed 32 out of 32 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
tests/merge_hand_e2e.rs	End-to-end test covering MergeHand epic→main lifecycle invariants.
tests/assign_close_cli.rs	CLI integration test for mandatory chat-of-record posting gate on assign close.
src/workshop.rs	Adds per-tab IRC view state and helpers to open/focus IRC projection tabs.
src/screen/mod.rs	Registers the new irc_view screen module.
src/screen/bench.rs	Adds TabKind::IrcView and a menu action to open the IRC view.
src/panel_state.rs	Re-exports IrcViewState via panel_state façade.
src/mcp/mod.rs	Introduces MCP registry/dispatcher and tool descriptors.
src/mcp/chat_tools.rs	Implements chat.post/chat.tail MCP tools with JSON schema + tests.
src/main.rs	Wires the new mcp module into the bin crate.
src/hand_archetypes.rs	Adds MergeHand archetype id + write-capable tool policy + tests.
src/cli.rs	Adds post, channel tail, mcp list/call, and assign close gate plumbing.
src/app.rs	Routes IRC view messages, refresh scheduling, and tab focus refresh behavior.
ipc/tests/chat_of_record.rs	Integration tests for post_message, tail, and count_posts_since_claim.
ipc/tests/channel_projection.rs	Integration tests for projection axes, FTS, mentions override, presets + unread.
ipc/tests/atlas_channel.rs	Integration tests for #atlas persistence + boot/shutdown seat logging.
ipc/src/lifecycle.rs	Boots #atlas channel + seat logging and persists shutdown marker.
ipc/src/lib.rs	Exposes new channel_projection and chat_of_record modules.
ipc/src/chat_of_record.rs	Adds durable post/tail primitives and mandatory-post counting helper.
ipc/src/channel_projection.rs	Implements filtered message projection, presets CRUD, and unread counts.
ipc/src/channel_manager.rs	Adds #atlas channel constants + sentinel spark creation/join helpers.
ipc/Cargo.toml	Adds uuid dependency for v4 ids.
docs/HEAD_ARCHETYPES.md	Documents chat-of-record boundaries and the tool-gated close behavior.
data/tests/transition_phase.rs	Adds tests for conflict handoff + merge-only “Merged” transition behavior.
data/tests/github_applier.rs	Adjusts lifecycle expectations due to new MergeHand-only transition edge.
data/src/sparks/transition.rs	Adds conflict-handoff rule, reason validation, and override lockdown flags.
data/src/sparks/error.rs	Adds InvalidReason transition error for reason-gated edges.
data/migrations/023_projection_presets.sql	Adds projection_presets table + indexes for saved projection filters.
RYVE.md	Documents chat-of-record workflow and CLI usage.
Cargo.toml	Enables sqlx migrate feature for #[sqlx::test(migrations=...)] usage.

[Copilot]

detect_active_seat_claim only inspects the last TAIL_MAX_LIMIT rows of #atlas via chat_of_record::tail. Since #atlas is intended to be a long-lived chat-of-record channel (boot + routing + handoffs), it can easily exceed 1000 messages; if an unreleased online, seat claim marker scrolls out of this window, a new Atlas could incorrectly boot as Claim instead of Follower. Consider querying only the seat-marker rows (e.g. raw_text LIKE 'atlas % online, seat %' OR 'atlas % offline, seat released') and scanning all of those, or doing a targeted SQL existence check for an unreleased claim without a hard tail cap.

[daxxsyntax]

Addressed in 2d3846a — real bug. Switched from chat_of_record::tail(... TAIL_MAX_LIMIT) (last 1000 rows of ALL #atlas traffic) to a filtered SQL query that only reads rows starting with 'atlas ' (the shape the parsers recognise), bounded to 5000 rows. For plausible workshop lifetimes that covers every seat event in history; the bound is a belt-and-braces guard.

[Copilot]

The migration comment states unread count is derivable as max(irc_messages.id matching preset filters) - last_seen_message_id, but ipc::channel_projection::preset_unread_count explicitly avoids that arithmetic (uses COUNT(*) WHERE id > last_seen) because ids aren’t gap-free within a filter slice. Please update this comment to match the actual unread-count contract so future readers don’t reintroduce the incorrect assumption.

Suggested change

	-- had observed while this preset was active. Unread count is derivable as
	-- `max(irc_messages.id matching preset filters) - last_seen_message_id`
	-- without touching this table beyond a read.
	-- had observed while this preset was active. Unread count is derived by
	-- counting `irc_messages` rows that both match the preset filters and
	-- have `id > last_seen_message_id`; it is not safe to subtract IDs
	-- because matching message IDs need not be contiguous within a filtered
	-- projection.

[daxxsyntax]

Not fixed in this PR — data/migrations/*.sql files are sqlx migrations tracked by checksum. Even a comment-only edit breaks the migration tracker on every existing workshop (house rule documented in user memory). Accurate description lives in ipc::channel_projection::preset_unread_count's doc, which is what callers actually read. We could ship an 024_*.sql corrigendum migration if the outdated comment on 023 causes real confusion, but for now the source-of-truth doc in the source file is the right pointer.

[Copilot]

Module docs say IRC wire delivery is “best-effort … via the outbox relay”, but post_message never enqueues an outbox row and there’s no other code path here that sends the message to the IRC server. Either implement actual best-effort wire emission (e.g. via outbox or an injected IrcClient) or update the docs to state clearly that this module only persists to irc_messages and wire delivery (if any) is handled elsewhere.

Suggested change

	//! primary key. The database write is the durability contract; IRC wire
	//! delivery via the outbox relay is best-effort and may no-op when no
	//! relay is running.
	//! primary key. The database write is the durability contract provided
	//! by this module; IRC wire delivery, if any, is handled elsewhere and
	//! is not performed here.

[daxxsyntax]

Addressed in 2d3846a — doc reworded. post_message writes the durable row and nothing else; cross-agent coordination happens through ryve channel tail reading the DB. IRC-wire emission for chat-of-record posts is flagged as a 0.4.0 candidate. Prior wording ("best-effort via outbox relay") was aspirational and inconsistent with the implementation.