Catalog
Every x402, MPP, MCP, and skill listing we've indexed, normalized, and enriched.
cache-sweep-activation-fallback
Sweep stale extension caches before activation, then fall back to a sticky manual-refresh flag if the live refresh fails.
cache-safe-stop-hook-snapshotting
Persist a cache-safe stop-hook context snapshot only for main session queries so later helpers can resume from stable state without fork pollution.
cache-prefix-stability-for-subagents
Preserve cache-critical request fields when a child agent must share the parent's prompt prefix for cache hits.
cache-identical-fork-prefix-construction
Build forked conversation prefixes that stay identical across turns so prompt caching keeps working.
cache-edit-pin-and-replay
Queue cache edits for the next microcompact request, pin them to user-message positions, replay pinned edits later, and reset state only on explicit lifecycle boundaries.
bundled-skill-file-extraction
Lazily extract referenced files for bundled skills so large prompts stay offline until needed.
bundled-and-mcp-turn-zero-skill-filter
For first-turn skill listings, prefer a small bundled+MCP subset and fall back to bundled-only when the filtered set is still too large.
built-in-agent-feature-gating
Gate built-in agents based on feature flags and entrypoint types so SDK sessions stay lean.
budget-diminishing-continue
Issue continuation nudges while the turn stays under budget, then halt once diminishing returns or the budget cap justify stopping.
budget-aware-skill-listing-formatting
Derive the active model context budget at runtime and format skill listings to fit it instead of emitting raw command sets.
bridge-safe-command-advertising
Advertise only slash commands that a bridge client can actually invoke by filtering announcements with the same bridge-safety predicate used at execution time.
bridge-origin-safe-slash-override
Keep remote slash-command skipping enabled by default, then reopen it only for bridge-safe commands while denying unsafe local commands with a friendly local response.
background-marketplace-reconciliation
Run marketplace diff and reconciliation in the background while surfacing per-source progress to the UI without blocking startup.
autocompact-recursion-guards
Block autocompact from running in recursive contexts so forks and helpers don’t deadlock.
autocompact-circuit-breaker
Break the autocompact retry loop after several consecutive failures to avoid hammering shared APIs.
auto-refresh-after-new-plugin-sources
Trigger a plugin refresh cycle whenever reconciliation installs new marketplaces so plugin state stays in sync.
auto-memory-tool-permission-fencing
Gate every tool call in the auto-memory extract agent so only safe read-only actions and targeted writes execute.
atomic-post-await-state-commit
After awaiting persistence work, update the shared budget state for each `tool_use_id` exactly once so no concurrent reader ever sees a half-committed decision.
append-only-sidechain-transcripts
Persist every subagent transcript entry by appending with explicit parent UUIDs so transcripts become a linear, resumable sidechain.
api-round-ptl-retry-truncation
On prompt-too-long retries, drop the oldest API-round groups first and inject a synthetic user preamble when needed so the summarize request stays valid.
api-round-message-grouping
Split conversation history into API-round buckets so compaction retries cut only the chunks tied to the failed round.
ansi-safe-border-caption-embedding
Embed captions into terminal borders while preserving alignment, clipping, and color/dim semantics.