activity-tester

Installation
SKILL.md

Activity Tester

Test one activity through real business paths. obligations.json owns expectations, evidence and verdicts. The Agent designs/adjudicates; the runner persists/validates; an integrated host continues fresh batches. Reports remain available throughout.

Scope And Evidence

  • LLM selection belongs to the caller/executor, not this Skill. Inherit its configured model unless the user explicitly requests an override; never pin a model from historical evaluation results.
  • Execute only in env=test with current installed act-* Skills and act-cli. Apply the selected act_id boundary to requirements. Related IDs are integration inputs unless explicitly included.
  • Current delivery scope is requirements, reward delivery (including attached assets), and messages, with their necessary boundary/repeat/side-effect checks. ThinkingData/server-telemetry execution is deferred until its integration is ready and explicitly enabled. Do not invoke it, create executable telemetry obligations, or claim telemetry coverage; disclose required telemetry as currently excluded. Existing telemetry records remain readable, not automatically passed or erased. Client/H5/frontend telemetry remains outside scope.
  • Admit expectations from in-scope requirements, released reachable configuration or selected-activity runtime evidence. Compiled handlers/registered routes alone do not create requirements. Separate a broken prerequisite from unexecuted downstream behavior.
  • Require exact owning-carrier observations. HTTP success, response reward fields, static wiring and restoration do not prove reward delivery. Confirm every defect independently.
  • Only runner commands modify the ledgers. operations.json records physical attempts, never business results. Never replay STARTED/UNKNOWN: reconcile readback first. Restore temporary shared mutations before unrelated work.

Start Or Resume

Locate scripts/activity_tester.py as <tester> and scripts/activity_test_host.py as <host> relative to this Skill. Do not read their source or all references at startup.

Installs
2
First Seen
May 11, 2026
activity-tester from mercury-api.wepieoa.com