feat(cli): rehome cline-cli and codex throughput, with the schema sweep and decode tests it needs - #940
Open
ozymandiashh wants to merge 3 commits into
Conversation
ozymandiashh
force-pushed
the
fix/rehome-new-files
branch
from
August 5, 2026 13:19
5bec4a6 to
ebdb9f8
Compare
ozymandiashh
marked this pull request as ready for review
August 5, 2026 13:39
ozymandiashh
marked this pull request as draft
August 5, 2026 14:10
…ayout Upstream added two files under the old `src/` path that have no counterpart here: the cline-cli provider (448d470, with its follow-up c642787) and tool-excluded active throughput metrics for Codex (a6bf81f). Files added on one side only produce no merge conflict. When main is merged into this branch, git will happily create both at `src/…` — a path npm workspaces does not build and no registry imports. Green build, and a whole provider quietly does not exist. Nobody gets a conflict marker to notice. cline-cli follows the phase-8 recipe: discovery and I/O host-side, pure record decode in core, registered alongside its neighbours. The throughput port needed two repairs the original did not, both of which would have left the dashboard's Tok/s column empty. First, the write hop: the mapper that converts provider calls into cached turns dropped activeDurationMs, activeGeneratedTokens and toolWaitMs, so the three fields never survived the round-trip. Fixed in providerCallToCachedCall, with the read-back in cachedCallToApiCall carrying them too. Second, append-resume: a parse that ends mid-task (the live-session case) cached the task's calls with no timing, and when the appended task_complete arrived the decoder's per-pass timing window had already been discarded — the fields stayed missing until a full re-parse. The decoder now threads the task window through its serialized state (taskResultStart / taskGeneratedTokens / taskToolIntervals / taskStartedAt / openToolStarts), and a task_complete that closes a window opened in an earlier pass returns a timing patch the CLI applies to the concatenated prior+new calls. CODEX_CACHE_VERSION 10 and the codex session-cache parse version bump once so already-cached sessions re-derive. Both links are covered end-to-end rather than by mapper unit tests that would pass while the column stayed empty: a parseAllSessions round-trip test (cold + warm), an end-to-end append case through parseAllSessions, a CLI resume test asserting the mid-task-cut append equals a cold decode of the grown file, and a core split-invariant test that the timing survives ANY two-pass split with the serialized state threaded.
ozymandiashh
force-pushed
the
fix/rehome-new-files
branch
from
August 5, 2026 14:20
ebdb9f8 to
1159744
Compare
ozymandiashh
marked this pull request as ready for review
August 5, 2026 14:40
… keys Three contract defects in the published observation surface, and a fourth that the first attempt at fixing them created. **`model` and `pricingModel` were unbounded free text** — `z.string().min(1)`, no max, no pattern — inside a schema whose docstring states that no field can hold free text. For vercel-gateway the value arrives from a fetched report. The first attempt bounded them with a pattern. Review proved that wrong: real providers emit display names, not identifiers. Antigravity reads `payload.model.display_name` directly, Warp's alias map is a closed ten-entry list that returns anything unmapped verbatim, and devin emits the same shape. A pattern would have rejected the envelope for those providers, and the next Warp model to ship would have broken decoding — a breaking narrowing of a contract already published at 0.9.20. So the value is normalized at the observation boundary instead, which is what this package already does for tool names, rather than validated and rejected. **The smuggling test asserted the leak survives** — it planted a prompt in `model` and checked it was still there. Inverting that assertion exposed the real carrier: `dedupKey` is built from the model string, so bounding the model while the key is derived from it is half a fix. The secret rode through the key. **`sourceRef` was documented as an opaque fingerprint while every caller passed a raw absolute path**, and five decoders folded it straight into dedup keys — which are observation output. A reviewer produced a schema-valid envelope containing a real home directory path. Rather than rewriting the contract text to match the leak, the decoders now fingerprint it: codebuff, lingtai-tui, zerostack, pi/omp and grok all route through the keyed helper. That also matters for getagentseal#931, which derives sync span ids from these keys. **Cross-PR review: the fingerprint was keyed NOWHERE on the production path.** The first version of this change added `sourceRefFingerprint` with an unkeyed SHA-256 fallback, justified by the CLI bridge's old comment that the rich decoder "never consumes" the privacy key. That comment was overtaken by this very change — the decoders consume the key precisely via the dedup keys being fingerprinted here, so the fallback shipped dictionary-attackable digests of absolute paths in the same dedup keys, one layer down from the raw-path leak it replaced. The bridge now threads the host privacy key (getHostPrivacyKey, per-install stable, already used by the optimize detectors) into the rich decode, and `sourceRefFingerprint` now requires a key and throws on an empty one (decision D1, like every other fingerprint in the module) — an empty key can no longer silently degrade to an unkeyed digest. The bridge parity goldens derive their expected dedup keys under the same host key, and the session / daily cache bumps below still fire exactly once: the only released key shape is the raw-path one, so the re-parse drops those keys and lands the keyed fingerprint shape in the same pass. The codebuff golden pinned the raw-path form. It is updated with the derivation shown, and a comment recording that the raw path was the defect — the same trap the kiro golden set in a sibling PR.
Carries the regression tests from getagentseal#928, which is closed in favour of this branch. That PR reduced to tests alone once review established its source fixes were already here — and the tests are the part this branch was missing: it restored the behaviour without anything proving it stays restored. Each pins one restoration and fails against a decoder without it: custom tool calls reaching the turn, the eight-type fork-replay skip in both directions, the Buffer path synthesizing info and invocation, and exec mapping to Bash.
ozymandiashh
changed the title
Collaborator
Author
|
This PR is now stacked, and absorbs #928. Three commits:
Green as it stands: core 545, targeted CLI 69. Previously it was red on its own and would have stayed red until #933 landed — which is not a PR anyone should be asked to merge. |
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 join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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.
Upstream added two files under the old
src/path that have no counterpart on this branch: the cline-cli provider (448d470, plus its follow-upc642787) and tool-excluded active throughput metrics for Codex (a6bf81f).Why this is not just a missing feature
Files added on one side only produce no merge conflict. When main is merged into this branch, git will happily create both at
src/…— a path npm workspaces does not build and no registry imports. Green build, no conflict marker, and a whole provider quietly does not exist.That is the failure mode this PR exists to prevent, and it is invisible by construction.
What landed
cline-cli follows the phase-8 recipe: discovery and I/O host-side, pure record decode in core, registered alongside its neighbours, with the deduped-rollup follow-up folded in.
codex-throughput needed one repair the original did not. The mapper that writes provider calls into the session cache dropped
activeDurationMs,activeGeneratedTokensandtoolWaitMs, so the three fields never survived the round-trip — and the dashboard's Tok/s column could not have populated regardless of what was ported.Worth being explicit that the consumer is real: the header, the
anyActiveTiminggate, the calculation and the model-breakdown aggregation are all present on this branch. This is not a column ported into a void; the only broken link was the write hop.The test goes end-to-end through
parseAllSessionsand asserts the exact shape the aggregation consumes. A mapper unit test would have passed while the column stayed empty — which is precisely how the gap survived the original port.Verification
Core 521. Seven targeted CLI files, 50 tests, covering the cache round-trip, throughput, the TPS surface, codex providers, rich capture and cache invalidation.