Rigorous Research Agent
A validation, recovery, and source-tracking protocol for multi-step research.
The skill requires tool-failure logging, source traceability, cross-source validation, and limitation reporting, which reduces silent errors and fabricated citations, so it earns partial trust credit. It does not specify least privilege, user confirmation, sensitive-data handling, isolation of external effects, rollback, or dependency security; its requirement for detailed reasoning blocks may also expose unnecessary internal information, so points are deducted.
The instruction-level workflow is internally coherent and covers source ranking, error recovery, completion checks, and read-back verification, making the happy path plausible and meriting a moderate static score. However, it provides no executable implementation, standardized tool interface, concrete exception taxonomy, or reproducible key-path tests; failure feedback still depends on the executing agent, so points are deducted.
Activation conditions are reasonably specific, covering multi-source research, three or more tool interactions, file validation, and completeness requirements. Non-fit boundaries and input/output contracts are not defined, the referenced search/read_url/fetch tools are generic rather than environment-neutral, and there is no Chinese-language or mainland-China reachability guidance, so points are deducted.
The SKILL.md has clear sections for activation, concepts, anti-patterns, practices, guidelines, examples, and limitations; repository context supplies an MIT license and CI signals. The skill itself lacks stable versioning, changelog, maintenance ownership, update path, installation/dependency notes, and FAQs; generated optimization metadata is not governance evidence, so points are deducted.
Validation checkpoints, explicit error acknowledgment, source tracking, cross-source comparison, and a pre-completion checklist are directly useful for improving research workflows, supporting a near-static-cap score. There is no standardized final-output format, executable template, or direct evidence that this skill's key paths produce usable results; mandatory detailed reasoning also adds cost and implementation burden, so points are deducted.
The skill contains concrete rules, and its references include an optimization summary and documented failure patterns, providing limited auditability. These materials are primarily generated records and author-side analysis, not independent reproduction of the skill's key paths; the supplied tests target other components and there is no third-party corroboration, so only a low score is justified.
- This is a static-only review; actual behavior in a target agent, tool set, or filesystem is unconfirmed.
- Do not assume the host supports thinking blocks, source-tracking tables, or read_file verification without mapping them to available tools.
- Detailed reasoning traces may contain sensitive or internal information; prefer auditable summaries, sources, failures, and decision records.
- Repository optimization scores and research claims should not be treated as independent test evidence for this skill.
What does this skill do, and when should you use it?
Rigorous Research Agent is an operating protocol for multi-step research workflows. It addresses tool failures, source validation, incomplete evidence, opaque reasoning, and premature conclusions. The protocol adds validation checkpoints, explicit failure recovery, source traceability, structured reasoning blocks, and cross-source comparison. It is a good fit for agent systems that need auditable research behavior rather than simple answer generation.
It instructs an agent to evaluate sources before reading them, validate tool outputs, acknowledge failures, select recovery strategies, track retrieved and failed URLs, compare evidence across sources, and run a pre-completion checklist. It requires mapping sources to supported claims and documenting contradictions or remaining gaps. The supplied SKILL.md contains no scripts and does not specify a concrete tool integration or output file.
- A researcher consolidating findings from several sources can use it to assess relevance, completeness, and consistency.
- An agent engineer building workflows with search, URL reading, or fetch operations can use it to define failure handling and fallback behavior.
- A team producing auditable reports can use it to distinguish successfully retrieved evidence from prior references or unverified claims.
- A workflow that saves, writes, or reads research files can use it to require content-level verification after file operations.
What are this skill's strengths and limitations?
- Covers source evaluation, error recovery, evidence traceability, and completion validation.
- Requires failures and information gaps to be recorded instead of silently ignored.
- Useful for multi-source research and auditable agent workflows.
- The supplied frontmatter contains only name and description, with no visible platform-specific extensions.
- It is a workflow protocol, not an automated research tool; it does not search, fetch, or write files by itself.
- The source provides no scripts, test suite, or concrete tool integrations for this individual skill.
- Its structured reasoning requirements may add overhead to long tasks.
- The source provides no empirical cross-platform test results for this individual skill.
How do you install this skill?
Copy the directory containing SKILL.md into the project’s skills directory while preserving the folder layout. For example: mkdir -p .codex/skills && cp -R examples/interleaved-thinking/generated_skills/comprehensive-research-agent .codex/skills/. The README documents directory-based installation, but does not provide a dedicated installation command for this example skill.
How do you use this skill?
Activate it for a research request such as: “Research this topic, track every successfully retrieved source, acknowledge tool failures, cross-validate key claims, and list remaining gaps before concluding.” It is intended for workflows involving multiple sources, explicit verification requirements, file validation, or three or more tool interactions.