Repository navigation
R4: Build Dhyanesh orchestration, simulation and release layer - #1
Merged
Merged
Conversation
Implement TinyAgent.run as a bounded, single-decision lifecycle. It uses Aisha's frozen types, registry and prompt contract; limits model calls to 1 + max_repairs; repairs from the original prompt; never retries backend errors or terminal tool failures; and preserves answer provenance and raw model outputs in AgentResult. Add deterministic SimulatedBackend, JSONL RecordingBackend and SHA-256 prompt-matched ReplayBackend. Add 35 ground-truth scenarios, the F01-F16 fault taxonomy, round-trip validation, repair and exhaustion variants, false-execution checks, and simulated/live/replay evaluation commands. Provide 50 labelled live queries and explicitly mark unrun measurements as "not executed" instead of publishing estimated accuracy. Add the public exports, CLI with a labelled simulated mode, examples, packaging, README, Python 3.10/3.12 CI, dependency and dynamic-execution AST gates, and Dhyanesh's integration tests. Ignore prd/ and generated artifacts so planning documents cannot enter this branch. This branch is based on Aisha's existing aisha branch. Her reflection, registry, prompt, contract files and tests remain unchanged. The new simulator passes their contract integration check. The full fault eval, live runtimes and CLI execution remain pending on the parser, executor, coercion and transport modules from the other owners; CI labels the simulated eval deferred until those modules exist. PRD-04 demo fixtures are kept separate because Aisha's search and set_alarm test signatures differ from the PRD-04 scenario contract. Verified: 31 unit tests pass; wheel builds with no runtime dependencies; staged diff passes whitespace checks; prd/ matches .gitignore.
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.
Implement TinyAgent.run as a bounded, single-decision lifecycle. It uses Aisha's frozen types, registry and prompt contract; limits model calls to 1 + max_repairs; repairs from the original prompt; never retries backend errors or terminal tool failures; and preserves answer provenance and raw model outputs in AgentResult.
Add deterministic SimulatedBackend, JSONL RecordingBackend and SHA-256 prompt-matched ReplayBackend. Add 35 ground-truth scenarios, the F01-F16 fault taxonomy, round-trip validation, repair and exhaustion variants, false-execution checks, and simulated/live/replay evaluation commands. Provide 50 labelled live queries and explicitly mark unrun measurements as "not executed" instead of publishing estimated accuracy.
Add the public exports, CLI with a labelled simulated mode, examples, packaging, README, Python 3.10/3.12 CI, dependency and dynamic-execution AST gates, and Dhyanesh's integration tests. Ignore prd/ and generated artifacts so planning documents cannot enter this branch.
This branch is based on Aisha's existing aisha branch. Her reflection, registry, prompt, contract files and tests remain unchanged. The new simulator passes their contract integration check. The full fault eval, live runtimes and CLI execution remain pending on the parser, executor, coercion and transport modules from the other owners; CI labels the simulated eval deferred until those modules exist. PRD-04 demo fixtures are kept separate because Aisha's search and set_alarm test signatures differ from the PRD-04 scenario contract.
Verified: 31 unit tests pass; wheel builds with no runtime dependencies; staged diff passes whitespace checks; prd/ matches .gitignore.