Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
docs(mcp-host): cite symbols instead of line numbers (#739)
Alfredo Lopez1ad2045ba415
diff --git a/mcp-host/package-lock.json b/mcp-host/package-lock.jsonindex 26665d124..c3449d463 100644--- a/mcp-host/package-lock.json+++ b/mcp-host/package-lock.json@@ -1,12 +1,12 @@{"name": "clerum-mcp-host",- "version": "0.1.425",+ "version": "0.1.426","lockfileVersion": 3,"requires": true,"packages": {"": {"name": "clerum-mcp-host",- "version": "0.1.425",+ "version": "0.1.426","license": "SEE LICENSE IN LICENSE","dependencies": {"@anthropic-ai/sdk": "^0.32.0",diff --git a/mcp-host/package.json b/mcp-host/package.jsonindex f3d1ea0b0..5142b8e21 100644--- a/mcp-host/package.json+++ b/mcp-host/package.json@@ -1,6 +1,6 @@{"name": "clerum-mcp-host",- "version": "0.1.425",+ "version": "0.1.426","description": "MCP Host - reads Host CRD and provides LLM access via OpenAI/Claude","main": "dist/main.js","engines": {diff --git a/mcp-host/src/core/extensions/__tests__/contextManager.antiThrash.test.ts b/mcp-host/src/core/extensions/__tests__/contextManager.antiThrash.test.tsindex bccce34be..1d0feef51 100644--- a/mcp-host/src/core/extensions/__tests__/contextManager.antiThrash.test.ts+++ b/mcp-host/src/core/extensions/__tests__/contextManager.antiThrash.test.ts@@ -5,8 +5,11 @@* plus the §8.3 Test A (defensive pending_approval guard does not mutate the* compaction counter).*- * Fixtures rely on the heuristic token estimator (`heuristicCount`): ceil(chars/4)+4- * per message (#731; it was floor(words×1.3)+4 before). We build deterministic+ * Fixtures rely on the heuristic token estimator (`heuristicCount`): per message,+ * ceil(B/4)+4, where B is the UTF-8 byte length of the content as it serializes+ * inside JSON (escapes included, quotes excluded), plus ceil(B/4) for each tool+ * call's JSON arguments (#731; it was floor(words×1.3)+4 before). For the+ * plain-ASCII fixtures below B equals the character count. We build deterministic* fixtures rather than mocking the estimator so the assertions exercise the real* wiring end-to-end.*/diff --git a/mcp-host/src/core/extensions/__tests__/contextManager.test.ts b/mcp-host/src/core/extensions/__tests__/contextManager.test.tsindex 7549c9b95..1eb7bb602 100644--- a/mcp-host/src/core/extensions/__tests__/contextManager.test.ts+++ b/mcp-host/src/core/extensions/__tests__/contextManager.test.ts@@ -465,8 +465,8 @@ describe('PressureContextManager - archived markdown', () => {* T-A3 — #731. The deployed decision path for `codex-subscription` and* `grok-subscription`: a `PressureContextManager` with NO token counter, so* `computeTokenPressure` falls through to `estimateTokens` → `heuristicCount`- * (`contextManager.ts:521-522`). Both providers map to `FallbackTokenCounter`- * (`llm/registryCore.ts:215-216`), which counts through the same+ * (via `tierDecisionTokens`). Both providers map to `FallbackTokenCounter`+ * (their `PROVIDERS` descriptors in `llm/registryCore.ts`), which counts through the same* `heuristicCount`, so this case covers the counter path as well. The* heuristic counts escaped UTF-8 bytes / 4, the quantity the contract caps;* the word count it replaced undercounted minified JSON and left this historydiff --git a/mcp-host/src/core/tokenizer/heuristic.ts b/mcp-host/src/core/tokenizer/heuristic.tsindex 3546cd664..93edeae3f 100644--- a/mcp-host/src/core/tokenizer/heuristic.ts+++ b/mcp-host/src/core/tokenizer/heuristic.ts@@ -23,7 +23,7 @@** This is the count that selects the compaction tier for EVERY provider in the* default configuration, not only a fallback: `CLERUM_TOKENIZER_DRYRUN`- * defaults to true (`config.ts:793`), and under dry-run `computePressure`+ * defaults to true (`tokenizerDryrun` in `config.ts`), and under dry-run `computePressure`* decides the tier from this heuristic and uses an exact counter only to record* the delta. It is exact nowhere — it ignores `contentParts` and* provider-specific framing.@@ -46,7 +46,7 @@ export function heuristicCount(messages: ChatMessage[]): number {total += Math.ceil(jsonStringBytes(msg.content ?? '') / 4) + 4// An assistant message that issues a tool call carries its payload here,// never in `content`; without this the whole call bills at the framing- // overhead alone (#731). `openaiTokenCounter.ts:58-62` walks the same field+ // overhead alone (#731). `OpenAITokenCounter.countSync` walks the same field// and also encodes each call's `id` and `name`; only the arguments are// counted here, since they carry the payload.for (const tc of msg.tool_calls ?? []) {diff --git a/mcp-host/src/llm/__tests__/codexSubscription.test.ts b/mcp-host/src/llm/__tests__/codexSubscription.test.tsindex 7eb10745e..c7fb920c0 100644--- a/mcp-host/src/llm/__tests__/codexSubscription.test.ts+++ b/mcp-host/src/llm/__tests__/codexSubscription.test.ts@@ -177,7 +177,7 @@ describe('CodexSubscriptionProvider', () => {const provider = new CodexSubscriptionProvider('gpt-5.3-codex', wired as never)// A single tool result over the byte cap. The message count is 4, far below// `maxMessages`, so the refusal can only come from the canonical-hash path- // at `codexSubscription.ts:339-352` - the one that measures real bytes.+ // (`hashCanonicalCodexRequest` in `execute`) - the one that measures real bytes.const oversized = [{ role: 'system' as const, content: 'you are a helpful assistant' },{ role: 'user' as const, content: 'list every contact' },@@ -401,8 +401,9 @@ describe('CodexSubscriptionProvider', () => {// any of the three patterns, so it only fails under a broadening wide// enough to swallow an unrelated field - the case T-C4 cannot see.// `maxOutputTokens` reaches the contract from the caller unclamped- // (`codexSubscription.ts:267-270` -> `:452`), so this is a refusal a caller- // can actually provoke, not a synthetic one.+ // (`completeSingleTurn` -> `execute` -> `buildRequest`, which copies+ // `max_tokens` into `generation.maxOutputTokens`), so this is a refusal a+ // caller can actually provoke, not a synthetic one.const history = [{ role: 'user' as const, content: 'summarize' }]const rejected = provider.completeSingleTurn(history, { max_tokens: 16_385 })diff --git a/mcp-host/src/llm/__tests__/grokSubscription.test.ts b/mcp-host/src/llm/__tests__/grokSubscription.test.tsindex 553dc2682..fafa49f77 100644--- a/mcp-host/src/llm/__tests__/grokSubscription.test.ts+++ b/mcp-host/src/llm/__tests__/grokSubscription.test.ts@@ -398,8 +398,8 @@ describe('GrokSubscriptionProvider', () => {const provider = new GrokSubscriptionProvider('grok-4.6', wired as never)// The Grok mirror of T-C. The guard above refuses on message count; this// history is 4 messages, so the refusal can only come from the- // canonical-hash path at `grokSubscription.ts:363-377` - the one that- // measures real bytes. #728 classified the message count correctly and left+ // canonical-hash path (`hashCanonicalGrokRequest` in `execute`) - the one+ // that measures real bytes. #728 classified the message count correctly and left// this path reporting `invalid_request`, which is what reached the user as// "Connection Error", a label that reads as transient.const oversized = [@@ -449,7 +449,7 @@ describe('GrokSubscriptionProvider', () => {const wired = deps()const provider = new GrokSubscriptionProvider('grok-4.6', wired as never)// The `messages[i].toolCalls` bound has no guard ahead of it in this file -- // unlike the message count, which `execute` refuses itself at `:352`. It can+ // unlike the message count, which `execute` refuses itself before hashing. It can// only be reached through the canonical hash, which makes it the one size// refusal whose classification depends entirely on the regex list. 257 is// the right number because #728 raised `maxToolCalls` to 256 (`8e12900e6`);@@ -493,8 +493,8 @@ describe('GrokSubscriptionProvider', () => {const provider = new GrokSubscriptionProvider('grok-4.6', wired as never)// `checkStructure` runs before `JSON.stringify`, so a structure with more// elements than the byte cap is refused by the element bound- // (`grok-provider-attempt-contract/index.cjs:214`) and never by the byte- // measurement at `:414`. This test is the runtime consumer of that message+ // (`checkStructure` in `grok-provider-attempt-contract/index.cjs`) and never+ // by the byte measurement in `parseGrokCompletionRequestV1`. This test is the runtime consumer of that message// rename: without it nothing on the Grok path observes the difference// between the element bound and the byte bound. The suffix is also why the// byte pattern is matched as a prefix - anchoring it at both ends would drop@@ -600,11 +600,12 @@ describe('GrokSubscriptionProvider', () => {// The negative half of the partition: a mutation that collapses the ternary// to `request_limit_exceeded` turns this test red.// `generation.maxOutputTokens is out of range`- // (`grok-provider-attempt-contract/index.cjs:376`) is a `limit` refusal that+ // (`parseGeneration` in `grok-provider-attempt-contract/index.cjs`) is a `limit` refusal that// shares no prefix with any of the three regexes, so it is the distant miss:// it survives a narrow widening and fails only under one broad enough to// swallow an unrelated field. `max_tokens` reaches the contract from the- // caller unclamped (`grokSubscription.ts:291-294` -> `:477`), so this is a+ // caller unclamped (`completeSingleTurn` -> `execute` -> `buildRequest`,+ // which copies `max_tokens` into `generation.maxOutputTokens`), so this is a// refusal a caller can provoke, not a synthetic one.const history = [{ role: 'user' as const, content: 'summarize' }]diff --git a/mcp-host/src/llm/codexSubscription.ts b/mcp-host/src/llm/codexSubscription.tsindex f70eb82a2..06536535b 100644--- a/mcp-host/src/llm/codexSubscription.ts+++ b/mcp-host/src/llm/codexSubscription.ts@@ -41,10 +41,12 @@ export type CodexAttemptContext = {* The contract `limit` refusals that mean "this turn carries too much".** Of the `fail('limit', …)` checks on a request, only these are about- * conversation volume: the real byte bound (`index.cjs:570`) and its- * non-image share on a V2 request (`:579`), the element bound that proxies it- * (`:332`), `maxMessages` (`:389`) and `messages[i].toolCalls` (`:418`).- * All four are "this conversation is too long", which is exactly what+ * conversation volume, all in `llm-provider-attempt-contract/index.cjs`: the+ * real byte bound and its non-image share on a V2 request (both in+ * `parseCodexCompletionRequestRoot`), the element bound that proxies it+ * (`checkStructure`), and `maxMessages` and `messages[i].toolCalls` (both in+ * `parseMessages`).+ * All five are "this conversation is too long", which is exactly what* `ContextLengthExceeded` — "Conversation Too Long" — promises the user.* Compaction reaches them unevenly. The context manager counts bytes and,* through the registry's `maxMessages`, the message count, so it compacts@@ -53,12 +55,14 @@ export type CodexAttemptContext = {* `maxToolCalls` also bounds every response, so only history produced by* another provider can carry an over-long `toolCalls` array.*- * The others are not. Nesting depth (`:313`, `:354`),- * `generation.maxOutputTokens` and `deadlineMs` out of range are malformed or+ * The others are not. Nesting depth (`checkStructure`, `assertFiniteTree`),+ * `generation.maxOutputTokens` (`parseGeneration`) and `deadlineMs`+ * (`parseCodexCompletionRequestRoot`) out of range are malformed or* out-of-range parameters, and a shorter conversation fixes none of them;* labelling them a context-length failure would send the user into a* compaction loop that cannot converge. They stay `invalid_request`, which is- * what `subscriptionRequestHash.test.ts:155-168` pins for the over-deep schema.+ * what "fails an over-deep tool schema locally without a stack overflow" in+ * `subscriptionRequestHash.test.ts` pins for the over-deep schema.* The image budgets belong to the attachments, not the conversation: a `size`* one is `attachment_too_large` (see `ATTACHMENT_BUDGET_REFUSALS`), the* `maxImages` `count` one stays `invalid_request`.diff --git a/mcp-host/src/llm/grokSubscription.ts b/mcp-host/src/llm/grokSubscription.tsindex 4e0058a8d..c024f1734 100644--- a/mcp-host/src/llm/grokSubscription.ts+++ b/mcp-host/src/llm/grokSubscription.ts@@ -36,17 +36,19 @@ export type GrokAttemptContext = {/*** The contract `limit` refusals that mean "this turn carries too much".*- * The Grok twin of `codexSubscription.ts:68-76`. The two are deliberate- * duplicates, not an accident: `grok-provider-attempt-contract/index.cjs:6`- * states that this package does not import the Codex contract's LIMITS, and a+ * The Grok twin of `CONTEXT_LENGTH_REFUSALS` in `codexSubscription.ts`. The two+ * are deliberate duplicates, not an accident: the header of+ * `grok-provider-attempt-contract/index.cjs` states that this package does+ * not import the Codex contract's LIMITS, and a* shared predicate would either recreate that coupling or pin prose rather than* code. The cost of the duplication is that each copy needs its own tests, which* is what T-C-grok, T-C2-grok, T-C3-grok and T-C5-grok are for. If you change* one list, read the other.** `fail('limit', …)` guards eight checks here too, and only these are about- * volume: the real byte bound (`index.cjs:414`), the element bound that proxies- * it (`:214`), `maxMessages` (`:260`) and `messages[i].toolCalls` (`:289`).+ * volume: the real byte bound (`parseGrokCompletionRequestV1`), the element+ * bound that proxies it (`checkStructure`), and `maxMessages` and+ * `messages[i].toolCalls` (both in `parseMessages`).* All four are "this conversation is too long", which is exactly what* `ContextLengthExceeded` — "Conversation Too Long" — promises the user.* Compaction reaches them unevenly. The context manager counts bytes and,@@ -56,13 +58,15 @@ export type GrokAttemptContext = {* `maxToolCalls` also bounds every response, so only history produced by* another provider can carry an over-long `toolCalls` array.*- * The other four are not. Nesting depth (`:199`, `:236`),- * `generation.maxOutputTokens` (`:376`) and `deadlineMs` (`:445`) out of range- * are malformed or out-of-range parameters, and a shorter conversation fixes- * none of them; labelling them a context-length failure would send the user into- * a compaction loop that cannot converge. They stay `invalid_request`, which is- * what `subscriptionRequestHash.test.ts:155-168` pins for the over-deep schema- * across both providers.+ * The other four are not. Nesting depth (`checkStructure`, `assertFiniteTree`),+ * `generation.maxOutputTokens` (`parseGeneration`) and `deadlineMs`+ * (`parseGrokCompletionRequestV1`) out of range are malformed or out-of-range+ * parameters, and a shorter conversation fixes none of them; labelling them a+ * context-length failure would send the user into a compaction loop that+ * cannot converge. They stay `invalid_request`, which is what "fails an+ * over-deep tool schema locally without a stack overflow" in+ * `subscriptionRequestHash.test.ts` pins for the over-deep schema across both+ * providers.** `hashCanonicalGrokRequest` returns `{ ok, code, message }` and nothing else,* so the message is the only discriminator available at this boundary (#731).