New:Introducing Socket Scanning for VS Code Marketplace Extensions.Learn more →
Get Started
a

argszero

npm

13.04K

total weekly downloads

Packages

Current Co-maintainers

Former Co-maintainers

96 packages

@argszero/cordis-plugin-aux-reasoning

Auxiliary-call reasoning policy for dsh: the harness marks its own background calls (compaction checkpoints, session titles) with a `purpose` and names no reasoning level for them, so `dsh-llm` fills the gap from the route's declared default and they thin

@argszero/cordis-plugin-balance-guard

Balance guard for dsh: reads GET /user/balance on the agent/pre-step waterfall and refuses the step when the provider says the account cannot pay for it (is_available=false, or the balance is at/below a configured floor). The refusal dispatches no model r

@argszero/cordis-plugin-client-bundle-conflict-guard

Client bundle conflict guard for dsh: the Web host concatenates the client bundles of every mounted plugin into one classic script per combo batch, with no per-bundle scope, so two bundles that declare the same name at their top level silently overwrite e

@argszero/cordis-plugin-compaction-route-fallback

Route fallback for a dsh compaction summary call the session's own window cannot fit: dsh-compaction-basic replays the shadowed region unbounded, so on a small-window route the call that must shrink the surface is rejected by it and the session self-locks

@argszero/cordis-plugin-content-exists-risk

Content-moderation rescue for dsh: a provider's content moderation (DeepSeek HTTP 400 'Content Exists Risk' / INVALID_REQUEST) rejects an entire request, and because the loop re-derives the whole conversation from the session log on every turn, one stored

@argszero/cordis-plugin-credential-rotate

Per-route API-key rotation + retry for dsh. When a provider request ends with a retryable failure (default: QUOTA, AUTH — the codes where waiting cannot help), this plugin (1) repoints the provider's credential reference (e.g. llm-deepseek's apiKeyEnv) to

@argszero/cordis-plugin-empty-response-guard

Guards for silent successful turns in dsh: observes the llm/stream waterfall and reclassifies (1) a reasoning-only completion (a stop with no text and no tool call) as EMPTY_RESPONSE and (2) a stop whose usage accounts for zero tokens on a route that has

@argszero/cordis-plugin-empty-response-retry

Makes an unattributed model attempt retryable, for the one seam that can tell. Three reports, one missing evidence. Discussion #8321: a `dsh-llm-pi-ai` route that relays an upstream empty response ends the turn, because `classifyPiAiError` has no rule mat

@argszero/cordis-plugin-fork-inbox-guard

Fork inbox guard for dsh: a created child that copies a prefix of another session's log replays that prefix's pending input into its own empty inbox, where it is addressed to the source session's next turn and runs ahead of the child's first message. Moun

@argszero/cordis-plugin-git-status

Read-only git workspace status chip for the dsh web composer (branch, ahead/behind, dirty counts) — reference implementation for the upstream #5073 feature request

@argszero/cordis-plugin-goal-ask-guard

Deny ask_user_question during autonomous goal rounds for the dsh harness: an agent-scoped ctx.tools.guard() on the agent/created + agent/inbox/claimed seam, as interim mitigation for #6074 (a nested ask during a goal run is not model-visible in code mode,

@argszero/cordis-plugin-goal-budget-guard

Deployment budgets for dsh autonomous goal rounds: cap a goal's rounds, the rounds that run with no human message, and the tokens spent since the guard began watching it. The harness bounds a goal only by its own `maxGoalRounds` (default 256), which is de

@argszero/cordis-plugin-goal-progress-guard

Progress judgement for dsh autonomous goal rounds: pause an active goal while its background jobs or delegated subagents are still running (resume when the last one drains), and stop a goal whose rounds have stopped producing tool calls. The harness drive

@argszero/cordis-plugin-history-sanitizer

Request-level history sanitizer for dsh: a session whose durable log holds content the target wire protocol cannot represent is permanently unusable, because the offending block sits in the log and no UI can edit it. Mounted on the public llm/stream water

@argszero/cordis-plugin-image-offload-fallback

Recover a session wedged by an image its route cannot serve: translate a decode rejection, or probe for an attachment object that is gone, into the harness's existing image-offload recovery

@argszero/cordis-plugin-inbox-claim-guard

Lost-input guard for dsh: a message claimed from the agent inbox is dropped if the step throws before the durable user/message append (discussion #6964, 14/154 archived sessions). This plugin observes the public agent/inbox/claimed event, proves the loss

@argszero/cordis-plugin-inbox-input-guard

Inbox input guard for dsh: a claimed inbox entry that is not a UserMessage crashes the turn inside whichever module reads it first, and the report names that reader - discussion #7363 ends ~1ms after turn/start, before any step/start, with "Cannot read pr

@argszero/cordis-plugin-inspect-client-guard

A platform:"client" cordis_inspect_query with no live page to answer it must not hang the asking agent forever. The shared Host registry (`CordisInspectRegistryService.queryClient`, packages/extensions/cordis-host-runner/src/inspect-registry.ts) settles i

@argszero/cordis-plugin-instruction-scope

Per-agent scoping for the dsh workspace instruction chain (discussion #7361). @deepseek-ai/dsh-agent-instructions injects the workspace instruction chain into every agent in the tree; its Config governs which files are discovered, globally, and nothing ex

@argszero/cordis-plugin-job-id-guidance

Job-id guidance for dsh: when job_output/job_kill fails with the registry's bare `unknown job <id>`, append one sentence naming the likely namespace mistake. Backgrounded continuable delegations return a durable subagent id and register no job, so the fai

@argszero/cordis-plugin-length-stop-overflow

Overflow guard for dsh: observes the llm/stream waterfall and restores the CONTEXT_WINDOW_EXCEEDED classification the built-in recovery keys on, for four ways a provider can refuse an over-long request without naming it. (1) A max-tokens finish whose outp

@argszero/cordis-plugin-live-model-listing

Ask a pi-ai provider endpoint what models it advertises, and show the difference against the installed catalog — a browser-half plugin for the dsh Models settings page

@argszero/cordis-plugin-llm-tool-call-guard

Runaway tool-call argument guard for dsh: wrap the llm/stream waterfall and cut a tool call that streams more than a configurable number of arguments bytes (UTF-8, maxArgsBytes) or fragments (maxArgsFragments), terminating with a routed error finish so th

@argszero/cordis-plugin-model-catalog-gap

Explains why a model id is unusable on a configured LLM provider route, by comparing the models this harness accepts for the route against the models the provider itself reports. The two lists come from different places and only one of them is readable of

@argszero/cordis-plugin-orphan-tool-call-guard

Orphan tool-call guard for dsh: a recorded tool call whose result was never durably recorded (a throwing tool scheduler `prepare`, a rejected dispatch, a cancelled turn) leaves an assistant tool_calls block the provider refuses on every later request, per