Correct context-lifecycle docs + document enforced per-fleet budgets#134
Merged
Conversation
The context-lifecycle page described a continuation-chain feature that was never built: the runtime re-queuing an exhausted run, a chain capping at 10, an 11th attempt labelled `chunk_chain_escalate_human`. None of those strings exist anywhere in the daemon or the runner. The runtime observes the context fill ratio and writes a log line; it cannot interrupt a fleet mid-loop, so it cannot force one to wrap up. A fleet that runs out of context finishes normally and the run ends `processed`. Nothing re-queues it. That page also answered "Will it run forever?" with three independent ceilings. All three were fiction: the context cap only logs, the continuation cap never existed, and `daily_dollars` was parsed and then ignored. It now names the ceilings that actually fire — the wall-clock deadline, the sandbox limits, the tenant credit gate, and the per-fleet budget that M122_005 makes real on the same branch. - context-lifecycle.mdx: rewritten. Continuation chains deleted; the observe-only threshold explained; the real terminal statuses and their labels tabulated. - authoring.mdx: the `<Note>` promising automatic re-enqueue replaced; the Budgets section now states both enforcement points, the two window shapes, that `monthly_dollars` is optional, and that a refused event is not charged. - troubleshooting.mdx: §5 distinguishes the pre-run refusal (`gate_blocked`) from the mid-run stop (`fleet_error`, `UZ-RUN-015`), and `budget_breach` from the credit-pool labels. - running.mdx, cli/agentsfleet.mdx: `continuation:<original_actor>` is an actor value nothing writes — removed from the activity-stream tags and the `--actor` filter examples. - changelog.mdx: a new Update. The shipped entry that carried the same claim is left intact — history is archived, not rewritten — and the correction is published forward instead. Every label, knob, and status named across the rewritten pages was checked against a non-test reader in src/. No phantoms remain. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Contributor
Author
|
Paired code PR: agentsfleet/agentsfleet#506 — enforces the per-fleet budget these pages describe. Merge alongside. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Paired with agentsfleet#506 (per-fleet budget enforcement + admin/models OpenAPI). Merge alongside it — these pages describe behaviour that PR enforces.
What changed
The context-lifecycle docs described a continuation-chain feature that was never built: the runtime re-queuing an exhausted run, a chain capping at 10, an 11th attempt labelled
chunk_chain_escalate_human. None of those strings exist anywhere in the daemon or the runner. The runtime observes the context fill ratio and writes a log line; it cannot interrupt a fleet mid-loop, so it cannot force one to wrap up. A fleet that runs out of context finishes normally and the run endsprocessed. Nothing re-queues it.That page also answered "Will it run forever?" with three ceilings — all fiction: the context cap only logs, the continuation cap never existed, and
daily_dollarswas parsed then ignored. It now names the ceilings that actually fire (wall-clock deadline, sandbox limits, tenant credit, and the per-fleet budget that agentsfleet#506 makes real).concepts/context-lifecycle.mdx— rewritten. Continuation chains deleted; the observe-only threshold explained; the real terminal statuses tabulated.fleets/authoring.mdx— the<Note>promising automatic re-enqueue replaced; the Budgets section now states both enforcement points, the two window shapes, thatmonthly_dollarsis optional, and that a refused event is not charged.fleets/troubleshooting.mdx— §5 distinguishes the pre-run refusal (gate_blocked) from the mid-run stop (fleet_error,UZ-RUN-015), andbudget_breachfrom the credit-pool labels.fleets/running.mdx,cli/agentsfleet.mdx—continuation:<original_actor>removed from the activity-stream tags and--actorexamples; no production code writes that actor value.changelog.mdx— a new<Update>. The shipped entry that carried the same continuation claim is left intact (history is archived, not rewritten); the correction is published forward.Every label, knob, and status named across the rewritten pages was checked against a non-test reader in the daemon's
src/. No phantoms remain.🤖 Generated with Claude Code
Greptile Summary
This PR corrects the docs for run lifecycle behavior and per-fleet budget handling.
Confidence Score: 5/5
This looks safe to merge.
Important Files Changed
Reviews (2): Last reviewed commit: "chore: merge main into served docs branc..." | Re-trigger Greptile