swarm: set the fresh-session threshold per agent
This commit is contained in:
parent
49ee678117
commit
5175d3663e
11 changed files with 154 additions and 32 deletions
|
|
@ -83,7 +83,7 @@ Manage agents across the swarm
|
||||||
* `mint-forge-token` — Check one agent's forge token, and mint it if it's missing or stale
|
* `mint-forge-token` — Check one agent's forge token, and mint it if it's missing or stale
|
||||||
* `set-state` — Declare the state a hive converges an agent to: `up`, `offline`, `paused` or `destroyed`
|
* `set-state` — Declare the state a hive converges an agent to: `up`, `offline`, `paused` or `destroyed`
|
||||||
* `watch` — Follow an agent's terminal rows live: the newest retained rows, then each new one, one line per row, until Ctrl-C or the stream ends
|
* `watch` — Follow an agent's terminal rows live: the newest retained rows, then each new one, one line per row, until Ctrl-C or the stream ends
|
||||||
* `inference` — Show the model, effort, subagent models, subagent cap and soft context limit the swarm sets for an agent, or change the model, effort and subagent models
|
* `inference` — Show the model, effort, subagent models, subagent cap and the two context limits the swarm sets for an agent, or change the model, effort and subagent models
|
||||||
* `permissions` — Show or change the permissions the swarm grants an agent
|
* `permissions` — Show or change the permissions the swarm grants an agent
|
||||||
* `agents-md` — Show an agent's AGENTS.md, or edit it with a diff
|
* `agents-md` — Show an agent's AGENTS.md, or edit it with a diff
|
||||||
|
|
||||||
|
|
@ -192,9 +192,9 @@ The swarm-level counterpart of `hivectl agent watch`, read from the swarm-contro
|
||||||
|
|
||||||
## `swarmctl agent inference`
|
## `swarmctl agent inference`
|
||||||
|
|
||||||
Show the model, effort, subagent models, subagent cap and soft context limit the swarm sets for an agent, or change the model, effort and subagent models.
|
Show the model, effort, subagent models, subagent cap and the two context limits the swarm sets for an agent, or change the model, effort and subagent models.
|
||||||
|
|
||||||
`get` prints the subagent cap (`max_subagents`) and the soft context limit (`soft_context_percent`); `set` has no flag for them and keeps stored ones. Only swarm-ui sets them.
|
`get` prints the subagent cap (`max_subagents`), the soft context limit (`soft_context_percent`) and the fresh-session threshold (`fresh_session_percent`); `set` has no flag for them and keeps stored ones. Only swarm-ui sets them.
|
||||||
|
|
||||||
A field the swarm sets wins over the agent's own value (its agent.nix or a `hyperhive-model`/`hyperhive-effort` file) from its next turn on; the agent reads its settings every ten seconds. A field the swarm doesn't set, or clears, leaves the agent on its own value.
|
A field the swarm sets wins over the agent's own value (its agent.nix or a `hyperhive-model`/`hyperhive-effort` file) from its next turn on; the agent reads its settings every ten seconds. A field the swarm doesn't set, or clears, leaves the agent on its own value.
|
||||||
|
|
||||||
|
|
|
||||||
|
|
@ -29,10 +29,17 @@ swarmctl agent inference clear iris
|
||||||
dialog; `set` has no flag for it either. It's the percent of the model's
|
dialog; `set` has no flag for it either. It's the percent of the model's
|
||||||
context window at which the agent compacts at the end of a turn, in place of
|
context window at which the agent compacts at the end of a turn, in place of
|
||||||
the harness's own 75%: 20 on a 1M window compacts once the context passes
|
the harness's own 75%: 20 on a 1M window compacts once the context passes
|
||||||
200k. A turn can still run past it, and it changes nothing else (the idle
|
200k. A turn can still run past it, and it changes nothing else (the
|
||||||
autoreset, the context shown in the UI). It's a whole number from 5 to 100;
|
context shown in the UI). It's a whole number from 5 to 100;
|
||||||
unset keeps the harness's own watermark
|
unset keeps the harness's own watermark
|
||||||
([`claude-invocation.md`](../turn-loop/claude-invocation.md)).
|
([`claude-invocation.md`](../turn-loop/claude-invocation.md)).
|
||||||
|
- You set the fresh-session threshold, `fresh_session_percent`, with it: the
|
||||||
|
percent of the context window above which a turn that finds the prompt cache
|
||||||
|
cold starts a fresh session instead of resuming, in place of the harness's
|
||||||
|
own 50%. It's a whole number from 5 to 100, and not above the soft limit when
|
||||||
|
both are set, since the agent compacts at the soft limit and the context would
|
||||||
|
never reach a higher threshold; the controller refuses a pair that crosses.
|
||||||
|
`set` has no flag for it either, and the dialog edits both on one slider.
|
||||||
- `get` shows each field, naming the unset ones (or says the swarm sets
|
- `get` shows each field, naming the unset ones (or says the swarm sets
|
||||||
nothing for the agent), and then the model and effort the agent reports
|
nothing for the agent), and then the model and effort the agent reports
|
||||||
running on, each with where it comes from.
|
running on, each with where it comes from.
|
||||||
|
|
@ -54,6 +61,7 @@ effort: (not set; the agent's own)
|
||||||
subagent_models: (not set; the agent's own)
|
subagent_models: (not set; the agent's own)
|
||||||
max_subagents: (not set; no cap)
|
max_subagents: (not set; no cap)
|
||||||
soft_context_percent: (not set; the harness's own watermark)
|
soft_context_percent: (not set; the harness's own watermark)
|
||||||
|
fresh_session_percent: (not set; the harness's own half)
|
||||||
running model: opus (swarm)
|
running model: opus (swarm)
|
||||||
running effort: medium (local_file)
|
running effort: medium (local_file)
|
||||||
```
|
```
|
||||||
|
|
@ -103,7 +111,8 @@ which swarm-controller alone writes:
|
||||||
"effort": "high",
|
"effort": "high",
|
||||||
"subagent_models": ["haiku", "sonnet"],
|
"subagent_models": ["haiku", "sonnet"],
|
||||||
"max_subagents": 3,
|
"max_subagents": 3,
|
||||||
"soft_context_percent": 20
|
"soft_context_percent": 20,
|
||||||
|
"fresh_session_percent": 10
|
||||||
}
|
}
|
||||||
```
|
```
|
||||||
|
|
||||||
|
|
@ -130,7 +139,9 @@ in-memory model or effort without writing `hyperhive-model` or
|
||||||
|
|
||||||
The same loop hands `soft_context_percent` to the Bus
|
The same loop hands `soft_context_percent` to the Bus
|
||||||
(`set_soft_context_percent`), which the compaction policy reads when it decides
|
(`set_soft_context_percent`), which the compaction policy reads when it decides
|
||||||
at turn end; a key with no limit, or no key, clears it on the next read.
|
at turn end, and `fresh_session_percent` (`set_fresh_session_percent`), which
|
||||||
|
the pre-turn auto-reset reads; a key with no limit, or no key, clears it on the
|
||||||
|
next read.
|
||||||
|
|
||||||
The subagent daemon (`hive-subagent-mcp`'s `swarm_models`) reads
|
The subagent daemon (`hive-subagent-mcp`'s `swarm_models`) reads
|
||||||
`subagent_models` and `max_subagents` with the same direct get, under the
|
`subagent_models` and `max_subagents` with the same direct get, under the
|
||||||
|
|
|
||||||
|
|
@ -188,7 +188,8 @@ still applies.
|
||||||
- **Auto session-reset** — a third path (`turn::maybe_auto_reset`,
|
- **Auto session-reset** — a third path (`turn::maybe_auto_reset`,
|
||||||
pre-turn) that fires when both conditions hold: context is ≥ a watermark
|
pre-turn) that fires when both conditions hold: context is ≥ a watermark
|
||||||
(`HIVE_AUTO_RESET_WATERMARK_TOKENS`, default **50% of the reported
|
(`HIVE_AUTO_RESET_WATERMARK_TOKENS`, default **50% of the reported
|
||||||
window**, none until the API reports one) AND the time since the last turn
|
window**, none until the API reports one; the swarm replaces the 50% per
|
||||||
|
agent with its `fresh_session_percent`, [`inference.md`](../swarm/inference.md)) AND the time since the last turn
|
||||||
exceeds the assumed prompt-cache TTL (`HIVE_CACHE_TTL_SECS`, default
|
exceeds the assumed prompt-cache TTL (`HIVE_CACHE_TTL_SECS`, default
|
||||||
`3600`; `0` turns the check off, so an idle session is always resumed, and
|
`3600`; `0` turns the check off, so an idle session is always resumed, and
|
||||||
the harness logs a warning at start when it's off). Claude's prompt cache goes cold after a while; once it's cold,
|
the harness logs a warning at start when it's off). Claude's prompt cache goes cold after a while; once it's cold,
|
||||||
|
|
|
||||||
|
|
@ -218,6 +218,11 @@ pub struct Bus {
|
||||||
/// compaction policy compacts at it instead of the harness's own
|
/// compaction policy compacts at it instead of the harness's own
|
||||||
/// watermark. `None` until the swarm sets one.
|
/// watermark. `None` until the swarm sets one.
|
||||||
soft_context_percent: Arc<Mutex<Option<u8>>>,
|
soft_context_percent: Arc<Mutex<Option<u8>>>,
|
||||||
|
/// The swarm's fresh-session threshold, as a percent of the context window
|
||||||
|
/// (`fresh_session_percent` of the agent's `agent-inference` key): above it, a
|
||||||
|
/// turn that finds the prompt cache cold starts a fresh session. `None`
|
||||||
|
/// until the swarm sets one, which keeps the harness's own half.
|
||||||
|
fresh_session_percent: Arc<Mutex<Option<u8>>>,
|
||||||
/// Resolved model id from the most recent turn's `assistant` events
|
/// Resolved model id from the most recent turn's `assistant` events
|
||||||
/// (the API-echoed `message.model`, e.g. `claude-opus-4-8`), as
|
/// (the API-echoed `message.model`, e.g. `claude-opus-4-8`), as
|
||||||
/// opposed to the requested `--model` name which may be a short
|
/// opposed to the requested `--model` name which may be a short
|
||||||
|
|
@ -261,6 +266,7 @@ impl Bus {
|
||||||
last_turn_ended_unix: Arc::new(AtomicI64::new(0)),
|
last_turn_ended_unix: Arc::new(AtomicI64::new(0)),
|
||||||
api_context_window: Arc::new(Mutex::new(None)),
|
api_context_window: Arc::new(Mutex::new(None)),
|
||||||
soft_context_percent: Arc::new(Mutex::new(None)),
|
soft_context_percent: Arc::new(Mutex::new(None)),
|
||||||
|
fresh_session_percent: Arc::new(Mutex::new(None)),
|
||||||
last_resolved_model: Arc::new(Mutex::new(None)),
|
last_resolved_model: Arc::new(Mutex::new(None)),
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
@ -530,6 +536,17 @@ impl Bus {
|
||||||
*self.soft_context_percent.lock().unwrap()
|
*self.soft_context_percent.lock().unwrap()
|
||||||
}
|
}
|
||||||
|
|
||||||
|
/// Set (or clear, with `None`) the swarm's fresh-session threshold.
|
||||||
|
pub fn set_fresh_session_percent(&self, percent: Option<u8>) {
|
||||||
|
*self.fresh_session_percent.lock().unwrap() = percent;
|
||||||
|
}
|
||||||
|
|
||||||
|
/// The swarm's fresh-session threshold, as a percent of the window.
|
||||||
|
#[must_use]
|
||||||
|
pub fn fresh_session_percent(&self) -> Option<u8> {
|
||||||
|
*self.fresh_session_percent.lock().unwrap()
|
||||||
|
}
|
||||||
|
|
||||||
/// Walk a stream-json value for `tool_use` blocks and bump the
|
/// Walk a stream-json value for `tool_use` blocks and bump the
|
||||||
/// per-turn counter for each one we find. Called by the stdout
|
/// per-turn counter for each one we find. Called by the stdout
|
||||||
/// pump on every parsed line. Cheap when the line isn't an
|
/// pump on every parsed line. Cheap when the line isn't an
|
||||||
|
|
|
||||||
|
|
@ -21,7 +21,8 @@
|
||||||
//! start, the agent runs on its local values.
|
//! start, the agent runs on its local values.
|
||||||
//!
|
//!
|
||||||
//! `subagent_models` and `max_subagents` are not read here. `soft_context_percent`
|
//! `subagent_models` and `max_subagents` are not read here. `soft_context_percent`
|
||||||
//! is handed to the Bus, which the compaction policy reads at turn end.
|
//! is handed to the Bus, which the compaction policy reads at turn end, and so is
|
||||||
|
//! `fresh_session_percent`, which the pre-turn auto-reset reads.
|
||||||
|
|
||||||
use std::time::Duration;
|
use std::time::Duration;
|
||||||
|
|
||||||
|
|
@ -147,6 +148,9 @@ pub async fn run(bus: Bus) {
|
||||||
target
|
target
|
||||||
.bus
|
.bus
|
||||||
.set_soft_context_percent(read.soft_context_percent);
|
.set_soft_context_percent(read.soft_context_percent);
|
||||||
|
target
|
||||||
|
.bus
|
||||||
|
.set_fresh_session_percent(read.fresh_session_percent);
|
||||||
}
|
}
|
||||||
tokio::time::sleep(POLL_INTERVAL).await;
|
tokio::time::sleep(POLL_INTERVAL).await;
|
||||||
}
|
}
|
||||||
|
|
@ -234,6 +238,7 @@ mod tests {
|
||||||
subagent_models: None,
|
subagent_models: None,
|
||||||
max_subagents: None,
|
max_subagents: None,
|
||||||
soft_context_percent: None,
|
soft_context_percent: None,
|
||||||
|
fresh_session_percent: None,
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
|
||||||
|
|
@ -264,12 +264,16 @@ pub fn stall_sleep_secs() -> u64 {
|
||||||
|
|
||||||
/// Resolve the auto-reset watermark. Priority order:
|
/// Resolve the auto-reset watermark. Priority order:
|
||||||
/// 1. `HIVE_AUTO_RESET_WATERMARK_TOKENS` env var (explicit override).
|
/// 1. `HIVE_AUTO_RESET_WATERMARK_TOKENS` env var (explicit override).
|
||||||
/// 2. 50% of the window the API last reported, or `0` before one has been.
|
/// 2. The swarm's `fresh_session_percent` of the window the API last reported,
|
||||||
|
/// or 50% when the swarm sets none; `0` before a window has been reported.
|
||||||
///
|
///
|
||||||
/// `0` disables auto-reset entirely.
|
/// `0` disables auto-reset entirely.
|
||||||
fn auto_reset_watermark_tokens(bus: &Bus) -> u64 {
|
fn auto_reset_watermark_tokens(bus: &Bus) -> u64 {
|
||||||
env_u64("HIVE_AUTO_RESET_WATERMARK_TOKENS")
|
env_u64("HIVE_AUTO_RESET_WATERMARK_TOKENS").unwrap_or_else(|| {
|
||||||
.unwrap_or_else(|| bus.api_context_window().map_or(0, |window| window / 2))
|
let percent = u64::from(bus.fresh_session_percent().unwrap_or(50));
|
||||||
|
bus.api_context_window()
|
||||||
|
.map_or(0, |window| window * percent / 100)
|
||||||
|
})
|
||||||
}
|
}
|
||||||
|
|
||||||
/// Resolve the assumed cache TTL: `HIVE_CACHE_TTL_SECS` if set (`0` disables
|
/// Resolve the assumed cache TTL: `HIVE_CACHE_TTL_SECS` if set (`0` disables
|
||||||
|
|
@ -958,6 +962,11 @@ mod tests {
|
||||||
assert_eq!(auto_reset_watermark_tokens(&bus), 0);
|
assert_eq!(auto_reset_watermark_tokens(&bus), 0);
|
||||||
bus.set_api_context_window(1_000_000);
|
bus.set_api_context_window(1_000_000);
|
||||||
assert_eq!(auto_reset_watermark_tokens(&bus), 500_000);
|
assert_eq!(auto_reset_watermark_tokens(&bus), 500_000);
|
||||||
|
// The swarm sets the threshold: a percent of the same window.
|
||||||
|
bus.set_fresh_session_percent(Some(20));
|
||||||
|
assert_eq!(auto_reset_watermark_tokens(&bus), 200_000);
|
||||||
|
bus.set_fresh_session_percent(None);
|
||||||
|
assert_eq!(auto_reset_watermark_tokens(&bus), 500_000);
|
||||||
}
|
}
|
||||||
|
|
||||||
/// A turn that reports no window never triggers the proactive compaction,
|
/// A turn that reports no window never triggers the proactive compaction,
|
||||||
|
|
|
||||||
|
|
@ -42,6 +42,13 @@ pub struct InferenceSettings {
|
||||||
/// Absent keeps the harness's own watermark.
|
/// Absent keeps the harness's own watermark.
|
||||||
#[serde(default, skip_serializing_if = "Option::is_none")]
|
#[serde(default, skip_serializing_if = "Option::is_none")]
|
||||||
pub soft_context_percent: Option<u8>,
|
pub soft_context_percent: Option<u8>,
|
||||||
|
/// The percent of the model's context window above which a turn that finds
|
||||||
|
/// the prompt cache cold starts a fresh session instead of resuming, in the
|
||||||
|
/// same range as `soft_context_percent` and not above it when both are set
|
||||||
|
/// (the agent compacts at the soft limit, so the context would never reach a
|
||||||
|
/// higher threshold). Absent keeps the harness's own half.
|
||||||
|
#[serde(default, skip_serializing_if = "Option::is_none")]
|
||||||
|
pub fresh_session_percent: Option<u8>,
|
||||||
}
|
}
|
||||||
|
|
||||||
/// What `GET` answers for one agent.
|
/// What `GET` answers for one agent.
|
||||||
|
|
@ -87,8 +94,9 @@ mod tests {
|
||||||
subagent_models: Some(vec!["haiku".to_owned()]),
|
subagent_models: Some(vec!["haiku".to_owned()]),
|
||||||
max_subagents: Some(3),
|
max_subagents: Some(3),
|
||||||
soft_context_percent: Some(20),
|
soft_context_percent: Some(20),
|
||||||
|
fresh_session_percent: Some(10),
|
||||||
},
|
},
|
||||||
r#"{"model":"opus","effort":"xhigh","subagent_models":["haiku"],"max_subagents":3,"soft_context_percent":20}"#,
|
r#"{"model":"opus","effort":"xhigh","subagent_models":["haiku"],"max_subagents":3,"soft_context_percent":20,"fresh_session_percent":10}"#,
|
||||||
),
|
),
|
||||||
] {
|
] {
|
||||||
assert_eq!(serde_json::to_string(&value).unwrap(), wire);
|
assert_eq!(serde_json::to_string(&value).unwrap(), wire);
|
||||||
|
|
|
||||||
|
|
@ -49,6 +49,7 @@ fn settings_view(v: AgentInference) -> InferenceSettings {
|
||||||
subagent_models: v.subagent_models,
|
subagent_models: v.subagent_models,
|
||||||
max_subagents: v.max_subagents,
|
max_subagents: v.max_subagents,
|
||||||
soft_context_percent: v.soft_context_percent,
|
soft_context_percent: v.soft_context_percent,
|
||||||
|
fresh_session_percent: v.fresh_session_percent,
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
@ -100,12 +101,27 @@ pub(crate) fn into_stored(settings: InferenceSettings) -> Result<AgentInference,
|
||||||
"soft_context_percent is {percent}; it is from {MIN_SOFT_CONTEXT_PERCENT} to {MAX_SOFT_CONTEXT_PERCENT}, or left out for the harness's own watermark"
|
"soft_context_percent is {percent}; it is from {MIN_SOFT_CONTEXT_PERCENT} to {MAX_SOFT_CONTEXT_PERCENT}, or left out for the harness's own watermark"
|
||||||
));
|
));
|
||||||
}
|
}
|
||||||
|
if let Some(percent) = settings.fresh_session_percent {
|
||||||
|
if !(MIN_SOFT_CONTEXT_PERCENT..=MAX_SOFT_CONTEXT_PERCENT).contains(&percent) {
|
||||||
|
return Err(format!(
|
||||||
|
"fresh_session_percent is {percent}; it is from {MIN_SOFT_CONTEXT_PERCENT} to {MAX_SOFT_CONTEXT_PERCENT}, or left out for the harness's own half"
|
||||||
|
));
|
||||||
|
}
|
||||||
|
if let Some(soft) = settings.soft_context_percent
|
||||||
|
&& percent > soft
|
||||||
|
{
|
||||||
|
return Err(format!(
|
||||||
|
"fresh_session_percent is {percent}, above soft_context_percent {soft}; the agent compacts at the soft limit, so a fresh start above it would never happen"
|
||||||
|
));
|
||||||
|
}
|
||||||
|
}
|
||||||
Ok(AgentInference {
|
Ok(AgentInference {
|
||||||
model: settings.model,
|
model: settings.model,
|
||||||
effort: settings.effort,
|
effort: settings.effort,
|
||||||
subagent_models: settings.subagent_models,
|
subagent_models: settings.subagent_models,
|
||||||
max_subagents: settings.max_subagents,
|
max_subagents: settings.max_subagents,
|
||||||
soft_context_percent: settings.soft_context_percent,
|
soft_context_percent: settings.soft_context_percent,
|
||||||
|
fresh_session_percent: settings.fresh_session_percent,
|
||||||
})
|
})
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
@ -261,8 +277,8 @@ pub(crate) async fn get_agent_inference(
|
||||||
ones whole", body = InferenceSettings),
|
ones whole", body = InferenceSettings),
|
||||||
(status = 400, description = "not an identifier, not declared on any hive, or a \
|
(status = 400, description = "not an identifier, not declared on any hive, or a \
|
||||||
model name is empty, has whitespace, starts with `-`, or is listed \
|
model name is empty, has whitespace, starts with `-`, or is listed \
|
||||||
twice, or `max_subagents` is above 32, or `soft_context_percent` is outside \
|
twice, or `max_subagents` is above 32, or `soft_context_percent` or `fresh_session_percent` is outside \
|
||||||
its range (problem+json)",
|
its range, or the fresh-session percent is above the soft limit (problem+json)",
|
||||||
body = String),
|
body = String),
|
||||||
(status = 422, description = "an unknown field, or an effort that is not one of \
|
(status = 422, description = "an unknown field, or an effort that is not one of \
|
||||||
the five levels"),
|
the five levels"),
|
||||||
|
|
@ -406,6 +422,7 @@ mod tests {
|
||||||
subagent_models: Some(vec!["haiku".to_owned()]),
|
subagent_models: Some(vec!["haiku".to_owned()]),
|
||||||
max_subagents: None,
|
max_subagents: None,
|
||||||
soft_context_percent: None,
|
soft_context_percent: None,
|
||||||
|
fresh_session_percent: None,
|
||||||
}
|
}
|
||||||
);
|
);
|
||||||
assert_eq!(
|
assert_eq!(
|
||||||
|
|
@ -590,6 +607,43 @@ mod tests {
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
#[test]
|
||||||
|
fn fresh_session_percent_outside_its_range_is_refused() {
|
||||||
|
for percent in [
|
||||||
|
0,
|
||||||
|
MIN_SOFT_CONTEXT_PERCENT - 1,
|
||||||
|
MAX_SOFT_CONTEXT_PERCENT + 1,
|
||||||
|
] {
|
||||||
|
let json = format!(r#"{{"fresh_session_percent":{percent}}}"#);
|
||||||
|
assert!(into_stored(settings(&json).unwrap()).is_err(), "{json}");
|
||||||
|
}
|
||||||
|
assert!(settings(r#"{"fresh_session_percent":-1}"#).is_err());
|
||||||
|
}
|
||||||
|
|
||||||
|
#[test]
|
||||||
|
fn fresh_session_percent_in_range_is_stored() {
|
||||||
|
for percent in [MIN_SOFT_CONTEXT_PERCENT, 50, MAX_SOFT_CONTEXT_PERCENT] {
|
||||||
|
let json = format!(r#"{{"fresh_session_percent":{percent}}}"#);
|
||||||
|
let stored = into_stored(settings(&json).unwrap()).unwrap();
|
||||||
|
assert_eq!(stored.fresh_session_percent, Some(percent));
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
/// The agent compacts at the soft limit, so a fresh-session threshold above
|
||||||
|
/// it could never be reached.
|
||||||
|
#[test]
|
||||||
|
fn fresh_session_percent_above_the_soft_limit_is_refused() {
|
||||||
|
let above = r#"{"soft_context_percent":40,"fresh_session_percent":41}"#;
|
||||||
|
assert!(into_stored(settings(above).unwrap()).is_err());
|
||||||
|
for ok in [
|
||||||
|
r#"{"soft_context_percent":40,"fresh_session_percent":40}"#,
|
||||||
|
r#"{"soft_context_percent":40,"fresh_session_percent":10}"#,
|
||||||
|
r#"{"fresh_session_percent":90}"#,
|
||||||
|
] {
|
||||||
|
assert!(into_stored(settings(ok).unwrap()).is_ok(), "{ok}");
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
#[test]
|
#[test]
|
||||||
fn max_subagents_in_range_is_stored() {
|
fn max_subagents_in_range_is_stored() {
|
||||||
for max in [0, 1, 3, MAX_SUBAGENTS] {
|
for max in [0, 1, 3, MAX_SUBAGENTS] {
|
||||||
|
|
|
||||||
|
|
@ -54,6 +54,11 @@ pub struct AgentInference {
|
||||||
/// in place of the harness's own watermark. `None` keeps the harness's.
|
/// in place of the harness's own watermark. `None` keeps the harness's.
|
||||||
#[serde(default, skip_serializing_if = "Option::is_none")]
|
#[serde(default, skip_serializing_if = "Option::is_none")]
|
||||||
pub soft_context_percent: Option<u8>,
|
pub soft_context_percent: Option<u8>,
|
||||||
|
/// The percent of the model's context window above which a turn that finds
|
||||||
|
/// the prompt cache cold starts a fresh session instead of resuming. `None`
|
||||||
|
/// keeps the harness's own half.
|
||||||
|
#[serde(default, skip_serializing_if = "Option::is_none")]
|
||||||
|
pub fresh_session_percent: Option<u8>,
|
||||||
}
|
}
|
||||||
|
|
||||||
/// A claude `--effort` level, ascending.
|
/// A claude `--effort` level, ascending.
|
||||||
|
|
@ -147,11 +152,12 @@ mod tests {
|
||||||
subagent_models: Some(vec!["haiku".to_owned(), "sonnet".to_owned()]),
|
subagent_models: Some(vec!["haiku".to_owned(), "sonnet".to_owned()]),
|
||||||
max_subagents: Some(3),
|
max_subagents: Some(3),
|
||||||
soft_context_percent: Some(20),
|
soft_context_percent: Some(20),
|
||||||
|
fresh_session_percent: Some(10),
|
||||||
};
|
};
|
||||||
let encoded = serde_json::to_string(&value).unwrap();
|
let encoded = serde_json::to_string(&value).unwrap();
|
||||||
assert_eq!(
|
assert_eq!(
|
||||||
encoded,
|
encoded,
|
||||||
r#"{"model":"opus","effort":"xhigh","subagent_models":["haiku","sonnet"],"max_subagents":3,"soft_context_percent":20}"#
|
r#"{"model":"opus","effort":"xhigh","subagent_models":["haiku","sonnet"],"max_subagents":3,"soft_context_percent":20,"fresh_session_percent":10}"#
|
||||||
);
|
);
|
||||||
assert_eq!(
|
assert_eq!(
|
||||||
serde_json::from_str::<AgentInference>(&encoded).unwrap(),
|
serde_json::from_str::<AgentInference>(&encoded).unwrap(),
|
||||||
|
|
@ -176,8 +182,8 @@ mod tests {
|
||||||
);
|
);
|
||||||
}
|
}
|
||||||
|
|
||||||
/// A value written before `max_subagents` or `soft_context_percent`
|
/// A value written before `max_subagents`, `soft_context_percent` or
|
||||||
/// existed decodes with neither set.
|
/// `fresh_session_percent` existed decodes with none set.
|
||||||
#[test]
|
#[test]
|
||||||
fn a_value_without_the_caps_has_neither_set() {
|
fn a_value_without_the_caps_has_neither_set() {
|
||||||
let decoded: AgentInference =
|
let decoded: AgentInference =
|
||||||
|
|
@ -185,6 +191,7 @@ mod tests {
|
||||||
.unwrap();
|
.unwrap();
|
||||||
assert_eq!(decoded.max_subagents, None);
|
assert_eq!(decoded.max_subagents, None);
|
||||||
assert_eq!(decoded.soft_context_percent, None);
|
assert_eq!(decoded.soft_context_percent, None);
|
||||||
|
assert_eq!(decoded.fresh_session_percent, None);
|
||||||
assert_eq!(decoded.model.as_deref(), Some("opus"));
|
assert_eq!(decoded.model.as_deref(), Some("opus"));
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
|
||||||
|
|
@ -37,6 +37,7 @@ impl SetFlags {
|
||||||
subagent_models: self.subagent_models.or(current.subagent_models),
|
subagent_models: self.subagent_models.or(current.subagent_models),
|
||||||
max_subagents: current.max_subagents,
|
max_subagents: current.max_subagents,
|
||||||
soft_context_percent: current.soft_context_percent,
|
soft_context_percent: current.soft_context_percent,
|
||||||
|
fresh_session_percent: current.fresh_session_percent,
|
||||||
})
|
})
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
@ -131,7 +132,7 @@ fn render_configured(s: Option<&InferenceSettings>) -> String {
|
||||||
fn render(s: &InferenceSettings) -> String {
|
fn render(s: &InferenceSettings) -> String {
|
||||||
const UNSET: &str = "(not set; the agent's own)";
|
const UNSET: &str = "(not set; the agent's own)";
|
||||||
format!(
|
format!(
|
||||||
"model: {}\neffort: {}\nsubagent_models: {}\nmax_subagents: {}\nsoft_context_percent: {}\n",
|
"model: {}\neffort: {}\nsubagent_models: {}\nmax_subagents: {}\nsoft_context_percent: {}\nfresh_session_percent: {}\n",
|
||||||
s.model.as_deref().unwrap_or(UNSET),
|
s.model.as_deref().unwrap_or(UNSET),
|
||||||
s.effort.map_or(UNSET, Effort::as_str),
|
s.effort.map_or(UNSET, Effort::as_str),
|
||||||
s.subagent_models
|
s.subagent_models
|
||||||
|
|
@ -142,6 +143,10 @@ fn render(s: &InferenceSettings) -> String {
|
||||||
s.soft_context_percent.map_or_else(
|
s.soft_context_percent.map_or_else(
|
||||||
|| "(not set; the harness's own watermark)".to_owned(),
|
|| "(not set; the harness's own watermark)".to_owned(),
|
||||||
|n| format!("{n}%")
|
|n| format!("{n}%")
|
||||||
|
),
|
||||||
|
s.fresh_session_percent.map_or_else(
|
||||||
|
|| "(not set; the harness's own half)".to_owned(),
|
||||||
|
|n| format!("{n}%")
|
||||||
)
|
)
|
||||||
)
|
)
|
||||||
}
|
}
|
||||||
|
|
@ -203,12 +208,14 @@ mod tests {
|
||||||
assert_eq!(merged.effort, Some(Effort::High));
|
assert_eq!(merged.effort, Some(Effort::High));
|
||||||
}
|
}
|
||||||
|
|
||||||
/// `set` has no flag for the cap or the soft limit, and stored ones
|
/// `set` has no flag for the cap or the two context limits, and stored
|
||||||
/// survive it.
|
/// ones survive it.
|
||||||
#[test]
|
#[test]
|
||||||
fn set_keeps_the_stored_cap_and_soft_limit() {
|
fn set_keeps_the_stored_cap_and_context_limits() {
|
||||||
let current: InferenceSettings =
|
let current: InferenceSettings = serde_json::from_str(
|
||||||
serde_json::from_str(r#"{"max_subagents":3,"soft_context_percent":20}"#).unwrap();
|
r#"{"max_subagents":3,"soft_context_percent":20,"fresh_session_percent":10}"#,
|
||||||
|
)
|
||||||
|
.unwrap();
|
||||||
let merged = SetFlags {
|
let merged = SetFlags {
|
||||||
model: Some("opus".to_owned()),
|
model: Some("opus".to_owned()),
|
||||||
..flags()
|
..flags()
|
||||||
|
|
@ -217,6 +224,7 @@ mod tests {
|
||||||
.unwrap();
|
.unwrap();
|
||||||
assert_eq!(merged.max_subagents, Some(3));
|
assert_eq!(merged.max_subagents, Some(3));
|
||||||
assert_eq!(merged.soft_context_percent, Some(20));
|
assert_eq!(merged.soft_context_percent, Some(20));
|
||||||
|
assert_eq!(merged.fresh_session_percent, Some(10));
|
||||||
}
|
}
|
||||||
|
|
||||||
/// An effort the controller would answer 422 for fails before the request.
|
/// An effort the controller would answer 422 for fails before the request.
|
||||||
|
|
@ -267,16 +275,17 @@ mod tests {
|
||||||
serde_json::from_str(r#"{"configured":{},"effective":null}"#).unwrap();
|
serde_json::from_str(r#"{"configured":{},"effective":null}"#).unwrap();
|
||||||
assert_eq!(
|
assert_eq!(
|
||||||
render_configured(empty.configured.as_ref()).lines().count(),
|
render_configured(empty.configured.as_ref()).lines().count(),
|
||||||
5
|
6
|
||||||
);
|
);
|
||||||
}
|
}
|
||||||
|
|
||||||
#[test]
|
#[test]
|
||||||
fn an_empty_value_renders_every_field_as_unset() {
|
fn an_empty_value_renders_every_field_as_unset() {
|
||||||
let row = render(&serde_json::from_str("{}").unwrap());
|
let row = render(&serde_json::from_str("{}").unwrap());
|
||||||
assert_eq!(row.lines().count(), 5);
|
assert_eq!(row.lines().count(), 6);
|
||||||
assert!(
|
assert!(
|
||||||
row.ends_with("soft_context_percent: (not set; the harness's own watermark)\n"),
|
row.contains("soft_context_percent: (not set; the harness's own watermark)\n")
|
||||||
|
&& row.ends_with("fresh_session_percent: (not set; the harness's own half)\n"),
|
||||||
"{row}"
|
"{row}"
|
||||||
);
|
);
|
||||||
assert!(row.contains("max_subagents: (not set; no cap)\n"), "{row}");
|
assert!(row.contains("max_subagents: (not set; no cap)\n"), "{row}");
|
||||||
|
|
|
||||||
|
|
@ -386,13 +386,14 @@ enum AgentVerb {
|
||||||
/// show what it last did. Fails when the swarm has never heard from the
|
/// show what it last did. Fails when the swarm has never heard from the
|
||||||
/// agent.
|
/// agent.
|
||||||
Watch(AgentWatchArgs),
|
Watch(AgentWatchArgs),
|
||||||
/// Show the model, effort, subagent models, subagent cap and soft context
|
/// Show the model, effort, subagent models, subagent cap and the two context
|
||||||
/// limit the swarm sets for an agent, or change the model, effort and
|
/// limits the swarm sets for an agent, or change the model, effort and
|
||||||
/// subagent models.
|
/// subagent models.
|
||||||
///
|
///
|
||||||
/// `get` prints the subagent cap (`max_subagents`) and the soft context
|
/// `get` prints the subagent cap (`max_subagents`), the soft context limit
|
||||||
/// limit (`soft_context_percent`); `set` has no flag for them and keeps
|
/// (`soft_context_percent`) and the fresh-session threshold
|
||||||
/// stored ones. Only swarm-ui sets them.
|
/// (`fresh_session_percent`); `set` has no flag for them and keeps stored
|
||||||
|
/// ones. Only swarm-ui sets them.
|
||||||
///
|
///
|
||||||
/// A field the swarm sets wins over the agent's own value (its agent.nix
|
/// A field the swarm sets wins over the agent's own value (its agent.nix
|
||||||
/// or a `hyperhive-model`/`hyperhive-effort` file) from its next turn on;
|
/// or a `hyperhive-model`/`hyperhive-effort` file) from its next turn on;
|
||||||
|
|
|
||||||
Loading…
Reference in a new issue