Install any skill in seconds. Free to start, no credit card required.
Get Started Free →hashgraph-online/waggle-memory
Use Waggle automatically as portable, local project memory during meaningful coding work; retrieve prior decisions and constraints, then store only durable outcomes.
Passed 6 of 14 test cases with the skill, and 5 of 14 without it.
and 1 more in the full report
3 fixed7 passed either way11 still failing1 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 2 test cases the skill fixed. Same model, same prompt, with and without the skill.
When executing memory queries for `github.com/acme/search-service`, what agent identifier parameter must be included in the tool invocation?
| Checked | Without skill | With skill |
|---|---|---|
| Specifies agent_id parameter as codex | Missed | Passed |
The instructions your model reads when the skill loads.
Treat Waggle as memory owned by the project, not as a transcript archive or generic personal chat memory.
Before the first Waggle call in a task, choose one stable project value:
.git, such as github.com/acme/api.Reuse that value across sessions. Do not use only a directory basename because unrelated repositories can share it. Pass agent_id: "codex" and the current task/session identifier when one is available. Never mix remembered context from a different project into the answer.
prime_context once with the narrowest known scope.query_graph with the stable project scope. Start with max_nodes: 10, max_depth: 1, and retrieval_mode: "hybrid".get_related when a returned node ID needs graph context, and graph_diff when the user asks what changed.Use this threshold for every write:
Store information when forgetting it would likely cause duplicated work, a wrong future decision, or violation of an established constraint. Do not store something merely because it happened.
After a completed turn, call observe_conversation only when the turn crosses that threshold and contains durable project knowledge, such as:
Pass the actual completed user_message, the final assistant_response, and the same project, agent, and session scope used for retrieval.
test cases passed, vs 5 / 14 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.