panjose/Co-Scientist / skills/hypothesis-review-pipeline
hypothesis-review-pipeline
VerifiedRun the decomposed review pipeline for a single hypothesis and persist each review stage as a structured artifact.
hypothesis-review-pipeline
Goal:
- Run the decomposed review pipeline for a single hypothesis and persist each review stage as a structured artifact.
Inputs:
ResearchPlanMetaReviewHypothesis- optional resume context from
state/PIPELINE_STATE.json
Outputs:
hypotheses/<id>/REVIEW/INITIAL_REVIEW.jsonhypotheses/<id>/REVIEW/FULL_REVIEW.jsonhypotheses/<id>/REVIEW/DEEP_VERIFICATION.jsonhypotheses/<id>/REVIEW/OBSERVATION_REVIEW.jsonhypotheses/<id>/REVIEW/SIMULATION_REVIEW.jsonhypotheses/<id>/REVIEW/REVIEW_SUMMARY.jsonliterature/queries/<query_id>/*when full review or deep verification needs external literature evidence- updated
hypotheses/<id>/HYPOTHESIS.json - updated
state/PIPELINE_STATE.json - updated
state/CURRENT_STAGE.json
Sub-skills:
hypothesis-initial-reviewliterature-searchhypothesis-full-reviewhypothesis-deep-verificationhypothesis-observation-reviewhypothesis-simulation-reviewhypothesis-review-summary
Context Loading:
- Open
skills/shared-references/schema-index.md. - Read
packages/agent_contracts/review.pybefore writing any per-stage review artifact underhypotheses/<id>/REVIEW/. - Read
packages/agent_contracts/hypothesis.pybefore updating the embedded review payload insidehypotheses/<id>/HYPOTHESIS.json. - Read
packages/agent_contracts/literature.pyandskills/shared-references/literature-search-contract.mdbefore dispatchinghypothesis-full-revieworhypothesis-deep-verification. - Read
packages/agent_contracts/pipeline_runtime.pybefore updatingstate/PIPELINE_STATE.jsonorstate/CURRENT_STAGE.json. - Read
skills/shared-references/codex-reviewer-routing.mdbefore using any optional Codex reviewer subagent route. - Read
RUN_POLICY.yamlbefore dispatching review work so the activereview_rigorsetting is honored. - Read the current hypothesis, research plan, and any optional run-level review guidance before starting the stage sequence.
Execution Contract:
- This pipeline skill does not own hidden prompt templates.
- It orchestrates atomic review skills so the runtime can trace each review stage independently.
- Downstream atomic review skills remain responsible for the exact field-level contents of each standalone review artifact.
- The canonical
hypotheses/<id>/HYPOTHESIS.jsonartifact must be updated after every review stage so the embeddedreviewpayload stays in sync with the per-stage review files. - Use
from tools import sync_hypothesis_reviewafter each stage artifact write. The canonical synchronization helper is implemented inpackages/run_artifacts/review_sync.py; do not hand-edit embedded review JSON. - Use
from tools import sync_pipeline_stage_artifactsas the canonical paired write surface for entering theReflectionsubstage. - Treat
tools.sync_hypothesis_review(run_dir, hypothesis_id)as a hard gate. If it fails, stop the review sequence immediately instead of continuing with later stages. - On entry, call
tools.sync_pipeline_stage_artifacts(run_dir, current_phase="Reflection", current_skill="hypothesis-review-pipeline")before starting the review sequence. - When consumed by a host agent, validate both the embedded hypothesis review payload and the per-stage review artifacts. If validation fails, stop and repair the review bundle before running the next stage or returning control.
- Full review and deep verification must use
tools.search_literature(run_dir, request)or an existing matchingEvidenceBundleContractbefore making external literature support claims. - The review pipeline must not accept prompt-invented paper metadata or model-memory citations as formal evidence.
- The review pipeline must not accept empty completed review artifacts. A completed or passing stage must contain concrete, hypothesis-specific findings that downstream evolution and ranking can use.
- Do not use placeholder review phrases such as
Viable evolved hypothesis,Refined from parent,Must outperform parent,Synthesize evolved catalyst, orBenchmark against parentas the substantive content of any completed review artifact. - Codex reviewer subagents are optional. If unavailable, run the same review contract in the main thread and record
reviewerRoute = local_main_threadin the trace. - If a Codex reviewer subagent is used, record
state/agent_traces/codex/<skill>/<timestamp>.json; the trace is audit evidence and must not replace canonical review artifacts. - Reviewer subagents must not write deterministic mechanics artifacts or bypass
tools.sync_hypothesis_reviewand schema validation. - Regardless of
review_rigor, run the full review stack for every hypothesis:hypothesis-full-reviewhypothesis-deep-verificationhypothesis-observation-reviewhypothesis-simulation-review
review_rigormay shorten or deepen the reasoning inside those stages, but it must not disable any review artifact.
Execution Steps:
- Open
skills/shared-references/schema-index.mdandskills/shared-references/codex-reviewer-routing.md, then readpackages/agent_contracts/review.py,packages/agent_contracts/hypothesis.py,packages/agent_contracts/literature.py, andpackages/agent_contracts/pipeline_runtime.pybefore writing any per-stage review artifact, embedded review state, literature evidence artifact, reviewer trace, or run-level stage artifact. - Read the research plan, current hypothesis, optional meta-review guidance, and active
RUN_POLICY.yaml. - Before dispatching atomic review work, call
tools.sync_pipeline_stage_artifacts(run_dir, current_phase="Reflection", current_skill="hypothesis-review-pipeline"). - Run
hypothesis-initial-reviewand persist the resulting stage artifact. - Immediately call
tools.sync_hypothesis_review(run_dir, hypothesis_id)to synchronizereview.initial_reviewback intohypotheses/<id>/HYPOTHESIS.json. If sync fails, stop here and repair the artifact instead of continuing. - Determine the required reasoning depth from
review_rigor, then run the full remaining atomic review stack for the hypothesis. Whenhypothesis-full-revieworhypothesis-deep-verificationneeds external literature evidence, require those skills to calltools.search_literature(run_dir, request)or consume an existing matching evidence bundle. - After each later review stage, call
tools.sync_hypothesis_review(run_dir, hypothesis_id)again so the standalone stage artifact and aggregate hypothesis artifact remain consistent. If sync fails, stop before the next stage. - Run
hypothesis-review-summaryafter the required review stages finish. - Run
python -m tools.validation.contract_validation <run_dir> --skill hypothesis-review-pipeline. If validation fails, stop and repair the review bundle before declaring completion or returning control.
Completion Rule:
- This skill is complete only when all review-stage artifacts exist, the embedded review payload in
hypotheses/<id>/HYPOTHESIS.jsonmatches those stage artifacts, and the resulting files validate for downstream routing.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deployment | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy. | Stars: 119k | Updated | Claude CodeCodex | Verified | Install |
| childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |