Repository navigation
[pull] main from can1357:main - #107
Merged
Merged
Conversation
…prefix probe The block-scoped profiler named the freeze: 45 of the first 58 stalls in the user's session spent 60-80% of their time in #collectRegexSecretValues, under obfuscateProviderContext, u 8000 nder prepareProviderCall. Every request to the model rescans the entire conversation - hundreds of messages, 700 KB and growing - with twelve lookbehind credential regexes, three to four times over. The phase breadcrumbs (compose, emit, append) only named whatever happened to be open around it. Each built-in pattern now declares the literal prefixes every one of its matches contains (every alternative spelled out: gho_/ghp_/ghu_/ghs_/ghr_, xoxa-..xoxs-, sk_live_/sk_test_/rk_live_/rk_test_, ...), and the obfuscator skips the regex over text containing none of them. The probe honours the regex's case-insensitivity by folding the text once per scan. It is fail-safe: a regex without metadata - every custom secrets.yml pattern - always runs, and an empty list counts as no metadata. A test redacts a specimen of every alternative and case variant of every built-in pattern through both the gated and the unfiltered scan and requires byte-identical output. On a 716 KB, 700-message context, one request's obfuscation drops from 1018 ms to 19 ms. (cherry picked from commit 6afdb21)
…ilter perf(secrets): gate built-in credential regexes behind literal prefix probes
… discourage polling - Changed Bash tool formatBackgroundNotice to describe output injection and advise against polling. - Updated coding-agent Bash prompt to reflect new background notice semantics. - Adjusted event-controller async update test to use formatBackgroundNotice for expected result. - Modified bash rendering test to verify new notice text and absence of polling warning.
8000
- Dropped the `?q=<jq>` query form and the jq parser (`json-query.ts`); a slash path is now always JSON extraction, one segment per key or array index (`agent://Scout/reports/0/data`). - Removed the slash-as-hierarchy probe: nested subagent outputs are addressed only by their dotted id (`agent://Parent.Child`), so `agent://Parent/Child` unambiguously extracts the `Child` field. - Updated system prompt, task/async-result prompts, hub job headers, and docs to the new form.
…footer note - Expanded aux utils pattern to include all sha*sum utilities. - Removed the truncation footer note from the bash tool prompt.
…g flag - Added _yaml helper to render nested dicts as YAML for config generation. - Introduced _settings attribute and merged user-provided settings into the generated config.yml. - Extended CLI with a --setting flag and exported settings via OMP_BENCH_SETTINGS environment variable.
… CLI kind flag - Modified OpenRouter test mocks to compare URLs with inequality instead of suffix check. - Changed compatibility compile test to expect a `judge` node error instead of `video`. - Updated OpenAI Daybreak test to filter models by API and altered CLI kind flag test to use `hologram` instead of `video`.
The no-ffprobe branch asserted "requires ffprobe" but the VideoError message reads "Video operations require ffprobe". The branch only runs where ffprobe is absent (CI runners), so it never failed locally.
…edential The reserved-lane roster is issued per key, but the provider was wired like a public gateway: - `singularityApiModelManagerOptions` set no `cacheProviderId`, so the authoritative discovery cache lived in the bare `singularityapi` namespace. Switching keys — or pointing at a self-hosted proxy — kept serving the previous roster for the full 24h TTL, including ids the current key cannot call. It now hashes the resolved credential and the `/v1`-normalized endpoint through a shared `normalizeSingularityApiBaseUrl` (discovery passes a `/v1`-suffixed value while the registry passes the raw configured one, so both must normalize) and joins `CREDENTIAL_SCOPED_MODEL_CACHE_PROVIDERS`, which is what makes `ModelRegistry` hydrate it with the credential. - The lane's wire shape was scoped to two globs, so any other lane a key owns fell back to the openai-completions defaults (`max_completion_tokens`). `max-tokens-field` and `reasoning-content-field` move to provider scope; the effort ladder, disable mode, clamp and image declaration stay on the reviewed lane, because for a row whose ceiling discovery leaves unknown a clamp falls back to the OpenAI default ceiling. - The glob-vs-exact rationale was wrong: the compiler lowercases glob values (`compile-cascade.ts`) while exact selectors compare the raw id (`cascade.ts`). Reworded to the reason the globs are actually there — one lane ships as a prefixed id, a dated row and a short alias. Tests: the login paste now uses the controller's `fetch` seam instead of replacing `globalThis.fetch` file-wide, plus coverage for a rejected key, `isCatalogDescriptor` false, the unreviewed-lane wire shape, and the cache namespace across both call paths.
Adding the SingularityAPI rows replaced existing documentation instead of extending it: - `docs/environment-variables.md` swapped the `GITLAB_TOKEN` row out for `SINGULARITYAPI_API_KEY`; the GitLab providers are still documented in `docs/providers.md`. Restored, with the new row appended after `MODEL_API_KEY` / 8000 `META_API_KEY` like every other credential added later. - `docs/providers.md` deleted the `/login cloudflare-ai-gateway` paragraph while inserting the new table row. Restored. - The interactive API-key login list now names `/login singularityapi`.
singularityapi login flow
singularityapi main platform
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
See Commits and Changes for more details.
Created by
pull[bot] (v2.0.0-alpha.4)
Can you help keep this open source service alive? 💖 Please sponsor : )