How we write and verify listings
Last updated:
Sources and AI assistance
TaskToTool starts with source listings and public repository metadata. Our agents/Codex helped compile the candidate list, paraphrase descriptions and write editorial wording. Some text is copied. In the 24-listing provenance review of 22 Oct 2026, all 13 grokbot.dev titles and 10 summaries were verbatim source text under CC BY 4.0; one further summary was shortened. Two grokbot.dev summaries and two GitHub summaries use AI-written neutral replacements. Nine GitHub summaries paraphrase or shorten repository descriptions. GitHub titles identify repository names.
We credit grokbot.dev entries to their listed authors and grokbot.dev, link the content licence and identify changes. Linked posts and media retain their own rights. Workflow context, setup guidance, limitations, price caveats and difficulty wording use AI-assisted TaskToTool templates. Difficulty is inferred from hosted access versus a repository; it is not a test result.
The directory is static. Browsing it does not send prompts to an AI service, and our metadata refresh job makes no LLM calls.
Evidence labels
- Source-discovered
- We have an identifiable source record. Capability claims have not necessarily been checked against documentation. All 24 current listings have only this level of evidence.
- Doc-reviewed
- Requires a named human to substantively compare the listing’s material claims with documentation, recording the sources and revision, claims checked, discrepancies, limitations and review date. No current listing has this label.
- Hands-on-tested
- Requires documentation review plus a recorded practical test: named tester, version, environment, task, steps, expected and observed results, failures, limits and test date. The public claim must state the tested scope. No current listing has this label.
We will use the higher labels only after the corresponding work and evidence exist. API metadata, licence checks, opening a README and working links do not count as substantive human review. A higher label supports only its recorded scope; it is not a certification of every feature or of safety.
Dates and changes in evidence
The imported “Source checked” label dated 21 Oct 2026 meant an imported snapshot, not human verification. The saved refresh successfully observed all 24 sources on 21 Oct 2026 at 11:54 UTC. That was an automated source-metadata observation. None has a documented substantive human documentation review or hands-on test.
Listing evidence separates source activity, imported check dates and stored automated source-metadata observations. A verification date must identify whether it describes a source-metadata observation, documentation review or hands-on test. Missing evidence dates are “Not recorded.” Failed or deferred refreshes do not advance a successful observation date. Build dates, source activity, CSV creation, licence checks and link checks cannot stand in for a substantive human-review date.
Going forward, material claim or version changes require reassessment of any higher label. Missing, contradicted or out-of-scope evidence means a downgrade to the highest supported level, with a recorded reason and decision date. We retain the historical review or test record without presenting it as current; the surviving label keeps its own evidence date.
Responsibility and corrections
Bogdan Mihaila holds editorial responsibility. AI-assisted wording and source claims can be wrong. Check current documentation, capabilities, access and cost at the linked source before relying on a listing. We have no active paid placements or affiliate links; any future relationship must be clearly identified.
Send corrections to contact@tasktotool.net or follow our reporting process. See also Our method for source observations and link checks and Licensing and attribution.