From eb570c003d7025f8f07fb3036ee970541e5e5329 Mon Sep 17 00:00:00 2001 From: damocles Date: Sun, 2 Aug 2026 14:37:22 +0200 Subject: [PATCH] cap get_loose_ends todo output + add ack_todos_until bulk-clear (#2944) --- hive-agent-mcp/src/mcp/args.rs | 14 +++++++++ hive-agent-mcp/src/mcp/mod.rs | 49 ++++++++++++++++++++++++++------ hive-agent-mcp/src/mcp/render.rs | 42 +++++++++++++++++++++++++++ hive-agent-sock/src/lib.rs | 4 +++ hive-agent/prompts/system.md | 2 +- hive-agent/src/todo_server.rs | 16 +++++++++++ hive-agent/src/todos.rs | 49 ++++++++++++++++++++++++++++++++ 7 files changed, 167 insertions(+), 9 deletions(-) diff --git a/hive-agent-mcp/src/mcp/args.rs b/hive-agent-mcp/src/mcp/args.rs index 9f7e4e54..9cef4d41 100644 --- a/hive-agent-mcp/src/mcp/args.rs +++ b/hive-agent-mcp/src/mcp/args.rs @@ -41,6 +41,20 @@ pub struct AckUntilArgs { pub up_to: i64, } +/// MCP tool args for `ack_todos_until`. +#[derive(Debug, serde::Deserialize, schemars::JsonSchema)] +pub struct AckTodosUntilArgs { + /// Highest todo id to bulk-clear: every pending todo (bash-task + /// completions, matrix unread, forge activity) with `id <= up_to` is + /// acked in one call — the todo-store analogue of `ack_until` for the + /// message inbox. `get_loose_ends` suggests a value in its truncation + /// summary when there are more todos than it renders individually; + /// pass that value (or the highest id you've actually triaged) to + /// clear the backlog in one shot instead of cancelling ids one at a + /// time. + pub up_to: i64, +} + /// MCP tool args for `remind`. Exactly one of `delay_seconds` or /// `at_unix_timestamp` must be set; both / neither is a tool-side error. /// Hides the tagged `ReminderTiming` enum behind a flatter schema so the diff --git a/hive-agent-mcp/src/mcp/mod.rs b/hive-agent-mcp/src/mcp/mod.rs index dd78d313..5b4b053c 100644 --- a/hive-agent-mcp/src/mcp/mod.rs +++ b/hive-agent-mcp/src/mcp/mod.rs @@ -24,18 +24,18 @@ mod args; mod render; pub use args::{ - AckUntilArgs, AgentGetLooseEndsArgs, AnswerArgs, AskArgs, CancelLooseEndArgs, - CancelScheduleArgs, CreateRepoArgs, EditScheduleArgs, FireScheduleNowArgs, GetAgentMetaArgs, - GetHostJournalArgs, GetLogsArgs, KillArgs, RecvArgs, RemindArgs, RequestInitConfigArgs, - RequestSchedulePromptArgs, RestartArgs, SendArgs, SetStatusArgs, StartArgs, UpdateArgs, - UpdateMetaInputsArgs, + AckTodosUntilArgs, AckUntilArgs, AgentGetLooseEndsArgs, AnswerArgs, AskArgs, + CancelLooseEndArgs, CancelScheduleArgs, CreateRepoArgs, EditScheduleArgs, + FireScheduleNowArgs, GetAgentMetaArgs, GetHostJournalArgs, GetLogsArgs, KillArgs, RecvArgs, + RemindArgs, RequestInitConfigArgs, RequestSchedulePromptArgs, RestartArgs, SendArgs, + SetStatusArgs, StartArgs, UpdateArgs, UpdateMetaInputsArgs, }; pub use render::{annotate_retries, format_ack, format_agent_meta, format_recv}; use render::{ dial_agent_socket, format_matrix_summary, local_questions, local_reminders, local_todos, - loose_end_kind_label, mark_local_todo_done, matrix_unread_summary, parse_loose_end_kind, - render_loose_ends, reply_err, + loose_end_kind_label, mark_local_todo_done, mark_local_todos_done_until, + matrix_unread_summary, parse_loose_end_kind, render_loose_ends, reply_err, }; /// Write (or remove) the status file in the agent's own `state/` directory. @@ -347,7 +347,11 @@ impl AgentServer { at turn start to remember what you owe / what's owed to you without scrolling \ inbox history. Output is a short bulleted list with ids, ages in seconds, and \ the relevant context. Each `question` or `reminder` row can be cancelled by \ - passing its id + kind to `cancel_loose_end`. Empty result is reported clearly.\n\ + passing its id + kind to `cancel_loose_end`. Empty result is reported clearly. \ + Todos are capped at 40 rendered rows (newest first) so a large backlog never makes \ + this call fail — a trailer line reports how many more are pending and suggests an \ + `ack_todos_until` value to bulk-clear the rest in one call instead of triaging \ + hundreds of ids individually.\n\ Pass `agent: \"\"` to inspect a specific peer agent's threads. Direct \ child agents are always accessible. For non-children, the `query_agent_state` \ capability is required — without it the request is rejected with an error." @@ -547,6 +551,35 @@ impl AgentServer { .await } + #[tool( + description = "Bulk-clear local todos (loose-ends v2 — bash-task completions, matrix \ + unread, forge activity): every pending todo with `id <= up_to` is acked in one call, \ + same shape as `ack_until` for the message inbox. Use this when `get_loose_ends` \ + reports more todos than it renders individually (its truncation summary suggests a \ + value) or whenever a backlog has piled up past the point of clearing ids one at a \ + time — note the highest id you've actually triaged, or take the suggested value \ + verbatim to clear the whole shown-as-hidden tail. Only affects YOUR todos; ids above \ + `up_to` stay pending. Returns how many were newly acked." + )] + async fn ack_todos_until(&self, Parameters(args): Parameters) -> String { + let log = format!("{args:?}"); + run_tool_envelope("ack_todos_until", log, async move { + match mark_local_todos_done_until(args.up_to).await { + Some(hive_agent_sock::Response::Acked { count }) => { + format!("acked {count} todo(s) up to id {}", args.up_to) + } + Some(hive_agent_sock::Response::Err { message }) => { + format!("ack_todos_until failed: {message}") + } + Some(other) => format!("ack_todos_until unexpected response: {other:?}"), + None => "ack_todos_until: local todo socket unavailable \ + (HIVE_AGENT_SOCKET unset or harness unreachable)" + .to_owned(), + } + }) + .await + } + #[tool( description = "Create a git repo through hive-c0re. You CANNOT create repos with your \ own forge token (creation is disabled) — this is the only path. The repo is created in \ diff --git a/hive-agent-mcp/src/mcp/render.rs b/hive-agent-mcp/src/mcp/render.rs index 8a06d98d..5c37c6df 100644 --- a/hive-agent-mcp/src/mcp/render.rs +++ b/hive-agent-mcp/src/mcp/render.rs @@ -136,6 +136,21 @@ fn msg_id_tag(id: i64) -> String { } } +/// Hard cap on how many individual `Todo` lines `render_loose_ends` will +/// emit. Approvals/questions/reminders stay naturally bounded (they're +/// triaged interactively and don't self-multiply), but todos are pushed by +/// unattended producers (matrix/bash/forge) — an agent that goes a long +/// stretch without calling `get_loose_ends`, or whose producers outpace its +/// triage, can accumulate hundreds of them. Rendering all of them +/// unconditionally risks producing a tool result too large for the MCP +/// transport to return at all, which is worse than a bug: it's a deadlock +/// (the agent can't even see what's pending to start clearing it). Capping +/// here guarantees `get_loose_ends` always returns successfully; the +/// truncation summary tells the agent how to bulk-clear the rest via +/// `ack_todos_until` (see that tool's doc for why it's the intended escape +/// hatch, not a per-id triage loop). +const MAX_RENDERED_TODOS: usize = 40; + /// Inner renderer for a `Vec` already extracted from the socket /// reply. Called by the `get_loose_ends` handler, which injects the /// `UnreadMatrix` entry before formatting. @@ -145,7 +160,17 @@ pub(super) fn render_loose_ends(loose_ends: &[hive_sh4re::LooseEnd]) -> String { return "(no loose ends)".to_owned(); } let mut out = format!("{} loose end(s):\n", loose_ends.len()); + let mut shown_todos = 0usize; + let mut hidden_todos = 0usize; + let mut hidden_min_id: Option = None; for t in loose_ends { + if let hive_sh4re::LooseEnd::Todo { id, .. } = t + && shown_todos >= MAX_RENDERED_TODOS + { + hidden_todos += 1; + hidden_min_id = Some(hidden_min_id.map_or(*id, |m| m.min(*id))); + continue; + } match t { hive_sh4re::LooseEnd::Approval { id, @@ -233,9 +258,19 @@ pub(super) fn render_loose_ends(loose_ends: &[hive_sh4re::LooseEnd]) -> String { "- todo #{id} [{subsystem}{key}, {age_seconds}s old]: {summary}{src} \ (cancel_loose_end kind:\"todo\" id:{id} to clear)" ); + shown_todos += 1; } } } + if hidden_todos > 0 { + let min_id = hidden_min_id.unwrap_or(0); + let _ = writeln!( + out, + "- {hidden_todos} more todo(s) not shown (oldest not shown: #{min_id}) — \ + call ack_todos_until(up_to: {min_id}) to bulk-clear the old backlog, or \ + cancel_loose_end kind:\"todo\" id: to clear individually" + ); + } out } @@ -350,6 +385,13 @@ pub(super) async fn mark_local_todo_done(id: i64) -> Option Option { + dial_agent_socket(&hive_agent_sock::Request::MarkTodosDoneUntil { up_to }).await +} + /// Format a `Vec` into a per-room summary string. /// Single room / single message collapses to one line; multi-room /// expands to a bulleted list. Returns an empty string for empty input. diff --git a/hive-agent-sock/src/lib.rs b/hive-agent-sock/src/lib.rs index e3cd83ed..8e82dc52 100644 --- a/hive-agent-sock/src/lib.rs +++ b/hive-agent-sock/src/lib.rs @@ -59,6 +59,10 @@ pub enum Request { }, /// The agent marks one of its own todos done, by id. MarkTodoDone { id: i64 }, + /// The agent bulk-acks every un-acked todo with `id <= up_to` in one + /// shot — the todo-store analogue of the message inbox's `ack_until`, + /// for clearing a backlog that piled up past the point of per-id triage. + MarkTodosDoneUntil { up_to: i64 }, /// Schedule a reminder that fires into this agent's own turn loop at /// `timing` (harness-local — no broker round-trip). Same semantics as /// the old broker `Remind` request. `file_path`, when set, is where the diff --git a/hive-agent/prompts/system.md b/hive-agent/prompts/system.md index b00a7389..617efde1 100644 --- a/hive-agent/prompts/system.md +++ b/hive-agent/prompts/system.md @@ -2,7 +2,7 @@ You are hyperhive agent `{label}` (qualified: `{qualified_label}`){hive_identity Tools (hyperhive surface). Full signature + behavior for each comes from the tool's own MCP description (you already received it via the MCP tool schema) — this is just the map of what exists and which ones are gated, so you know where to look: -- **Inbox / messaging** (always available): `mcp__hyperhive__recv`, `mcp__hyperhive__ack_until`, `mcp__hyperhive__send`, `mcp__hyperhive__ask`, `mcp__hyperhive__answer`, `mcp__hyperhive__get_loose_ends`, `mcp__hyperhive__cancel_loose_end`, `mcp__hyperhive__remind`, `mcp__hyperhive__set_status`, `mcp__hyperhive__get_agent_meta`. Two habits worth internalizing beyond the tool descriptions themselves: prefer ending the turn over repeatedly polling `recv` when idle (only turn-boundaries observe in-container todo wakes — bash-task completions, matrix unread, forge activity — and ending the turn is also your checkpoint); and `ask`/`answer` are async — `ask` returns immediately with a question id, the reply lands later as a `question_answered` system event, never block a turn waiting on it inline. +- **Inbox / messaging** (always available): `mcp__hyperhive__recv`, `mcp__hyperhive__ack_until`, `mcp__hyperhive__send`, `mcp__hyperhive__ask`, `mcp__hyperhive__answer`, `mcp__hyperhive__get_loose_ends`, `mcp__hyperhive__cancel_loose_end`, `mcp__hyperhive__ack_todos_until`, `mcp__hyperhive__remind`, `mcp__hyperhive__set_status`, `mcp__hyperhive__get_agent_meta`. Two habits worth internalizing beyond the tool descriptions themselves: prefer ending the turn over repeatedly polling `recv` when idle (only turn-boundaries observe in-container todo wakes — bash-task completions, matrix unread, forge activity — and ending the turn is also your checkpoint); and `ask`/`answer` are async — `ask` returns immediately with a question id, the reply lands later as a `question_answered` system event, never block a turn waiting on it inline. If `get_loose_ends` reports a large todo backlog (it caps rendered rows at 40), don't try to triage hundreds of ids by hand — call `ack_todos_until` with the suggested cutoff to bulk-clear the stale tail in one shot. - **Extra MCP tools** (some agents only): `mcp____` — agent-specific (matrix client, scraper, db connector, etc.) declared in your `agent.nix` under `hyperhive.extraMcpServers`. First-class tools, already operator-approved at deploy time. - **Lifecycle** (_requires `lifecycle` tool group_, direct children only, no approval needed): `restart`, `kill`, `start`, `update`, `list_containers`. - **Approvals** (_requires `approvals` tool group_, queues an operator approval): `request_init_config`, `request_apply_commit`, `request_update_meta_inputs`. diff --git a/hive-agent/src/todo_server.rs b/hive-agent/src/todo_server.rs index c307e027..753a2a61 100644 --- a/hive-agent/src/todo_server.rs +++ b/hive-agent/src/todo_server.rs @@ -202,6 +202,7 @@ fn dispatch( } => clear_todo(store, &subsystem, key.as_deref(), all), Request::ListTodos { subsystem } => list_todos(store, subsystem.as_deref()), Request::MarkTodoDone { id } => mark_todo_done(store, id), + Request::MarkTodosDoneUntil { up_to } => mark_todos_done_until(store, up_to), Request::StoreReminder { message, timing, @@ -417,6 +418,21 @@ fn mark_todo_done(store: &Todos, id: i64) -> Response { } } +/// `MarkTodosDoneUntil` handler: bulk-acks every un-acked todo id at or +/// below `up_to` — the escape hatch for a backlog too large to triage +/// one-by-one (see `Todos::mark_done_until`'s doc for why it's needed). +fn mark_todos_done_until(store: &Todos, up_to: i64) -> Response { + match store.mark_done_until(up_to) { + Ok(count) => { + tracing::debug!(up_to, count, "todo bulk mark-done"); + Response::Acked { + count: u64::try_from(count).unwrap_or(0), + } + } + Err(e) => err(&e), + } +} + /// `Request::Compact` handler: gate on context usage, then queue the same /// deferred `compact_pending` flag the operator's `/compact` button sets. /// Mirrors `hive-agent::web_ui::actions::post_compact` but reachable from diff --git a/hive-agent/src/todos.rs b/hive-agent/src/todos.rs index 4fc08465..415b8656 100644 --- a/hive-agent/src/todos.rs +++ b/hive-agent/src/todos.rs @@ -245,6 +245,34 @@ impl Todos { Ok(n) } + /// Bulk-ack every un-acked todo with `id <= up_to` in one shot — the + /// same "note the highest id you've seen, clear everything up to it" + /// pattern the message inbox's `ack_until` already gives agents, applied + /// to this store. Exists for the case a per-id triage loop isn't worth + /// it: an agent that hasn't called `get_loose_ends` in a long stretch + /// (or one whose producers pile up faster than it triages) can end up + /// with a backlog large enough that the rendered list itself becomes + /// unwieldy — clearing it in one call is the escape hatch, same + /// `acked`-not-deleted semantics as [`Self::mark_done`] (a reconciled + /// producer's next `upsert` still sees the row to compare against). + /// Returns the number of rows newly acked. + /// + /// # Errors + /// + /// Propagates the sqlite update failure. + /// + /// # Panics + /// + /// Panics if the connection mutex is poisoned. + pub fn mark_done_until(&self, up_to: i64) -> Result { + let conn = self.conn.lock().unwrap(); + let n = conn.execute( + "UPDATE todos SET acked = 1, acked_at = ?1 WHERE acked = 0 AND id <= ?2", + params![Utc::now().timestamp(), up_to], + )?; + Ok(n) + } + /// List todos, newest-updated first. `subsystem = Some(..)` filters to /// one producer's set; `None` returns all. Excludes acked rows — once /// the agent has dismissed a todo it stays out of its own list, even @@ -476,6 +504,27 @@ mod tests { assert!(!s.has_any().unwrap(), "acked-only table reads as empty"); } + /// `mark_done_until` acks everything at-or-below the threshold id and + /// leaves later rows untouched, mirroring `ack_until`'s inbox semantics. + #[test] + fn mark_done_until_acks_up_to_threshold_only() { + let (_dir, s) = store(); + let (id1, _) = s.upsert("bash", None, "task 1", None).unwrap(); + let (id2, _) = s.upsert("bash", None, "task 2", None).unwrap(); + let (id3, _) = s.upsert("bash", None, "task 3", None).unwrap(); + assert_eq!( + s.mark_done_until(id2).unwrap(), + 2, + "acks id1 and id2, not id3" + ); + let left = s.list(None).unwrap(); + assert_eq!(left.len(), 1); + assert_eq!(left[0].id, id3); + // Idempotent: re-running over the same range acks nothing new. + assert_eq!(s.mark_done_until(id2).unwrap(), 0); + let _ = id1; + } + /// `reap_acked` only removes acked rows past the cutoff — a recent ack /// and any un-acked row both survive. #[test]