TheGoat395/Codex-Skills / skills/deep-research
deep-research
VerifiedResearch with scoped depth and counterevidence.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 5.3 KB |
| agents/openai.yaml | config | 229 B |
| references/access-and-continuity.md | markdown | 1.6 KB |
| references/challenger-and-handoff.md | markdown | 1.8 KB |
| references/coverage-and-research-passes.md | markdown | 2.6 KB |
| references/evidence-standard.md | markdown | 4.7 KB |
| references/research-delivery.md | markdown | 1.0 KB |
references/evidence-standard.md
View on GitHub# Evidence Standard
Use this standard to make research auditable, resumable, and proportionate. Adapt presentation to the task; do not force a large report when a short answer is sufficient.
## Claim Ledger
| Field | Required treatment |
|---|---|
| Claim | One atomic, falsifiable statement. |
| Decision relevance | Why it could affect the answer. |
| Source | Exact inspected source or record. |
| Authority | Governing, direct, original authoritative, independent analysis, reporting, experiential, or discovery-only. |
| Source date | Publication, effective, data, version, event, or commit date. |
| Access date | When the evidence was checked. |
| Scope | Applicable boundaries, definitions, units, exceptions, and environment. |
| Status | Verified fact, supported interpretation, inference, estimate, recommendation, disputed, or unknown. |
| Confidence | High, medium, or low with a concise reason when material. |
| Contradiction | Conflicting evidence and its resolution or unresolved state. |
| Freshness | Drift risk and the condition requiring re-verification. |
## Coverage Ledger
Derive coverage dimensions from the objective rather than a fixed list:
| Dimension | Why material | Expected evidence | Inspected coverage | Depth | Gap or next action |
|---|---|---|---|---|---|
Dimensions may represent questions, entities, artifact classes, alternatives, relationships, time periods, environments, evidence origins, or constraints. Report lightweight screening and deep inspection separately. Never imply equal depth across the entire mapped field.
## Candidate and Relationship Ledger
| Item | Identity evidence | Discovery path | Relevance | Deep-inspection decision | Reason |
|---|---|---|---|---|---|
For material relationships, retain:
| From | Relationship | To | Direct evidence | Inference | Confidence | Decision effect |
|---|---|---|---|---|---|---|
Do not convert co-occurrence, similarity, repetition, or speculation into a verified relationship.
## Access Attempt Ledger
| Target | Preferred route | Failure evidence | Alternate route | Result | Claim impact |
|---|---|---|---|---|---|
Use genuinely different authorized routes. Repeatedly issuing the same failed query does not count as exhaustion.
## Comparison Matrix
| Candidate | Criterion | Evidence | Judgment | Confidence | Caveat |
|---|---|---|---|---|---|
Define criteria and weights before scoring. Keep unknown distinct from poor performance. Run sensitivity checks when uncertain weights or assumptions could change the decision.
## Challenger Delta
| Initial conclusion | Challenge | New evidence | Changed or upheld | Decision effect |
|---|---|---|---|---|
The challenger pass must improve coverage, confidence, or the decision rather than repeat the first pass.
## Resumption Checkpoint
For long work, retain the minimum state required to resume accurately:
- objective and acceptance criteria;
- scope and governing sources;
- current thesis and confidence;
- completed coverage dimensions;
- decisive sources and claims;
- unresolved contradictions and access failures;
- next highest-value actions;
- temporary artifact locations, if any.
## Source Tests
- **Identity:** Is it the original evidence or a retelling?
- **Access:** Was the relevant body actually inspected?
- **Currency:** Is its date or version fit for the decision?
- **Scope:** Does it cover the claim's actual boundaries?
- **Method:** Are collection method, definitions, sample, and incentives visible?
- **Independence:** Do multiple sources share the same unverified origin?
- **Corroboration:** Is independent confirmation necessary and available?
- **Contradiction:** What credible evidence points elsewhere?
Treat links, titles, snippets, summaries, and inaccessible citations as discovery evidence only.
## Failure Checks
- Did the first plausible result anchor discovery?
- Did the coverage graph omit an adjacent branch that could change the decision?
- Did screening favor familiarity, popularity, or search rank over relevance?
- Were uncertain identities, relationships, or duplicated origins corrected?
- Were incentivized claims tested against observable or independent evidence?
- Were unstable facts verified at the necessary time and scope?
- Did any citation support only part of its attached claim?
- Was inference, absence of evidence, or failed access presented as fact?
- Were practical constraints and implementation economics separated from theoretical capability?
- Did a material access failure receive a genuinely different documented attempt?
- Did the challenger pass test the strongest countercase?
- Could the work resume accurately from its checkpoint after context loss?
- Does the coverage claim match the ledger rather than the effort expended?
Repair material failures or classify them explicitly before delivery.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| add-product-e2e-evalpaperclipai/paperclipAdd or extend a Paperclip full-stack runner E2E workflow, fixture, matcher, or report evidence path for local or Daytona execution. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| add-runner-evalpaperclipai/paperclipAdd or extend a Paperclip Runner protocol evaluation definition, roster, assertion, or report fixture with provenance and narrow validation. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| diagnose-why-work-stoppedpaperclipai/paperclipCompatibility entry point for failed, stalled, looping, or over-recovered Paperclip work. Use fix-failed-paperclip-run for the full diagnosis, regression, repair, real verification, and green-PR workflow; honor an explicitly requested diagnosis-only scope. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| fix-failed-paperclip-runpaperclipai/paperclipDiagnose and fix failed, stuck, slow, or incorrectly blocked Paperclip tasks and runs. Use with a task/run URL or failure report to trace the root cause, prove a regression red then green, check second-order effects, verify the real workflow, and deliver a PR with passing checks. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |