Repository navigation
Per-task failure tolerance: on_failure: continue (#23 C2) - #40
Open
minerva-sky wants to merge 4 commits into
Open
minerva-sky wants to merge 4 commits into
minerva-sky wants to merge 4 commits into
Conversation
TaskDefinition gains optional id, depends_on and needs; ExecutionPlan gains from_hash and validate! (duplicate ids, dangling references, self-dependency, cycles). All additive: flat plans serialize and load unchanged. First of three staged PRs for codenamev#20. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
TaskPlanner's structured-output schema now asks for a plan-local id,
a depends_on list and needs wiring on every task, and the prompt tells
the model what each is for. needs travels as {name, task} pairs since
strict schemas cannot describe a free-form object; the parser folds
them into the hash TaskDefinition stores. Malformed edges are dropped
with a warning. After parsing, the plan is run through
ExecutionPlan#validate! and any dangling reference or cycle is logged,
with the edges kept so the plan stays inspectable and editable.
Part 2 of 3 for codenamev#20 (stacked on codenamev#35).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…v#20, part 3 of 3) Both execution paths (`agentic plan --execute` and `agentic execute --plan`) now build tasks through one helper that resolves the plan-local ids in `depends_on` and `needs` to the Task objects it just built, then hands each task to `PlanOrchestrator#add_task` with its edges. Dependents wait for their upstream and receive its output under the declared name; flat plans still run as an unordered swarm, exactly as before. A plan whose graph cannot be scheduled (dangling id, cycle) fails fast with the validator's message rather than executing anyway. The non-quiet summary reports the edge count alongside the task count. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
TaskDefinition and PlanOrchestrator#add_task accept on_failure: (skip_dependents, the default, or continue). A continue task that fails for good no longer skips its dependents; they run and can branch on Task#succeeded? / Task#failure_of (and needs.succeeded? / failure_of for named inputs) while output_of stays nil. Retries still run first, intervention-required failures never continue, and the plan status stays :partial_failure. The planner schema asks the LLM for the flag and both CLI execution paths pass it through. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
minerva-sky
added
the
loop:self
Collaborator
Author
|
Diagnosis of the red Ruby 3.2.4 job: it isn't a nil / version split. Both failing logs show a dependency UUID coming back as |
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
1 participant
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.
Implements C2 from #23 (design in the research comment there). Stacked on #38, so the diff shows the #20 series until that merges; the new commit is
e2ff6f2.What
TaskDefinition#on_failure:"skip_dependents"(default) or"continue". Serialized only when non-default, so flat plans keep their shape; unknown values raise.PlanOrchestrator#add_task(..., on_failure:): same policy per task. When acontinuetask fails terminally, its dependents are scheduled instead of skipped. A dependency counts as met when it completed or its failure was tolerated.Task#succeeded?(dep)/Task#failure_of(dep)andNamedOutputs#succeeded?/#failure_of: dependents can branch on the failure.output_ofstaysnil, so tasks that ignore the flag see no change.on_failureenum plus one prompt instruction; unknown values from the LLM fall back to the default with a warning.build_task_graphnow yields a 4-tuple).Guard rails
continueonly applies once the task is still:failed.skip_dependentstask between a tolerated failure and its own dependents still skips downstream.overall_statusstays:partial_failure. Hiding a tolerated failure from the result would be lying to the journal.Tests
19 new examples (5 TaskDefinition, 2 NamedOutputs, 7 orchestrator, 1 planner, CLI expectations updated). 87/0 across the touched files; standardrb clean. CHANGELOG under Unreleased/Added. VERSION bump left to the maintainer.
Not in v1: tolerated-error predicates and a plan-level default, per the #23 note.
🤖 Generated with Claude Code