Ponytail: 9 agents execute the adapter, Cursor stays at text level
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisis ton rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
Discrepancies arise between the image projected by badges and popularity and the actual execution of Ponytail among code agents. Out of 22 listed integrations, only 9 launch an adapter; 13 are limited to text, including Cursor and 11 others with a simple file of 2,593 bytes. This flattening highlights that the "text level" does not provide several control mechanisms, while Claude Code is among the hosts that actually execute the adapter.
Badges and stars do not guarantee an executable integration
The number of stars awarded to Ponytail does not reflect the reality of the executed behavior: distributing effective functionality within the 2026 agent ecosystems requires more than just sharing an AGENTS.md file. Inconsistencies arise between the promises displayed by a badge and what is actually documented in the portability lists. The tiered model of Ponytail adapters is also compared to the practices of various agent harnesses for plugin management, revealing varied expectations behind the same mention of integration.
22 agents recorded, but only 9 execute the Ponytail adapter
Ponytail lists 22 coding agents. Among them, 9 actually execute an adapter, while 13 do not have one. Twelve of these 13 are limited to the same body of rules, a 2,593-byte AGENTS.md file or an identical copy, which represents 39% of the total actual rules without any of the five expected behaviors. Claude Code is among the hosts that truly execute the adapter, while Cursor and 11 others settle for the minimal file.
At the text level, five controls disappear; method and checks
The integration qualified as "text" does not include several major protections: intensity, stop, persistence, output contracts, and support for sub-agents; Cursor is among the involved hosts. To outline this panorama, the hosts were divided by a Python script into two groups, EXEC and TEXT, resulting in 9 and 13 respectively, while maintaining reservations when the boundary was blurred by installations or manifests. The distributed copies were checked through continuous integration and by re-hashing the rule files to verify the absence of drift. The Ponytail repository mainly includes benchmarks and tests, rather than simple adapter templates.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.