Remove navigation tool friction seen in the Flash smoke run

docs/GETTING-STARTED.md

@@ -272,7 +272,7 @@ stable checkout to bind.
 | Stale or unknown evidence / generation mismatch | Inspect status, refresh successfully, and use the newly returned generation |
 | Excluded or unsupported source | Inspect policy/exclusion metadata and use a bounded direct-read fallback; do not silently broaden scope |
 | Response exceeds budget | Select a smaller sufficient block or split the evidence request |
-| Packet already in progress | Wait for the active request to finish, then issue the next request sequentially |
+| Packet queue is full | Wait for queued packet requests to finish; up to eight wait behind the running one |
 | Git HEAD error | Select a Git checkout with a commit; packet provenance requires it |
 
 There is no automatic watcher, worktree rebinding or configuration-writing setup

docs/LOCAL-NAVIGATION.md

@@ -7,8 +7,10 @@ limited to 128 records per collection; this bridge does not change that format.
 Instead, an explicitly configured repository can have a disposable in-memory
 source index, independent of the signed evidence cache.
 
-The initial search contract is exact, case-insensitive ASCII identifier tokens
-on source lines. It is not semantic search, compiler-resolved relationships or
+The search contract is exact and case-insensitive on source lines: one ASCII
+identifier token, or a printable ASCII literal such as `local-source-unsigned`
+or `grants no authority` whose ends fall on whole tokens (so `source-unsig`
+matches nothing). It is not semantic search, compiler-resolved relationships or
 a replacement for signed evidence. Results must say `local-source-unsigned`.
 
 The engine retains source lines and their file hashes from an explicit refresh.
@@ -99,8 +101,9 @@ object.
 
 ## Check a draft answer's coverage
 
-`repository_coverage` is a deterministic pre-submit check. It takes up to eight
-`symbols` and the draft `answer` (at most 128 KiB), explores each symbol as
+`repository_coverage` is a deterministic pre-submit check. It checks up to eight
+distinct `symbols` per call (further ones, up to 64, are returned as not checked
+for a second call) and the draft `answer` (at most 128 KiB), explores each symbol as
 above and lists every definition, test, reference and importing file as
 `cited` (the repository-relative path appears in the answer), `named` (only its
 basename does) or `missing`, missing files first with their enclosing

docs/WORKER-PACKETS.md

@@ -69,7 +69,8 @@ without truncation. The tool rejects unavailable, stale or unknown navigation
 and a mismatched generation, and checks freshness again before returning. Refresh
 and rebuild after relevant changes. This is bounded freshness checking, not an
 atomic snapshot against hostile concurrent filesystem changes.
-Only one packet request runs at a time; a concurrent request fails immediately.
+Only one packet request runs at a time; concurrent requests wait in order (up to
+eight), and a further one fails immediately.
 Cancellation is checked around assembly and freshness inspection. The shared
 bounded assembler does not interrupt an individual file read or Git provenance
 check already in progress; cancelled results are not returned as successful packets.

docs/experiments/smoke-flash-20260923/RESULTS.md

@@ -0,0 +1,56 @@
+# Smoke run on a DeepSeek V4.1 Flash executor
+
+Locked protocol `three-way-deepseek-v4.1-flash-smoke-20260923`, run on
+23 September 2026 (23:49 to 00:04 UTC) with Claude Code 2.1.280, every executor
+on `deepseek-v4.1-flash:cloud` at medium effort through the local Ollama route,
+structured reviews on `claude-sonnet-5` at high effort. Two tasks, one
+repetition. This checks the repeated-run configuration end to end; it is not a
+comparison of arms.
+
+## Result: passed
+
+Every arm completed. Both code-change cells were accepted on the checker and
+the scope check with no reviewer session; the structured cells were reviewed
+against rubric version 2 with the review rules in the prompt, and every verdict
+parsed at the first attempt.
+
+| Task | Arm | Accepted | Rule | Executor input | Executor s |
+| --- | --- | --- | --- | ---: | ---: |
+| orientation, Context repo | plain | yes | checker and reviewer | 493,289 | 76 |
+| orientation, Context repo | graphify | no | checker and reviewer | 364,806 | 53 |
+| orientation, Context repo | context | no | checker and reviewer | 1,038,702 | 101 |
+| code change, Context repo | plain | yes | checker and scope | 908,372 | 179 |
+| code change, Context repo | graphify | yes | checker and scope | 267,330 | 49 |
+| code change, Context repo | context | yes | checker and scope | 560,266 | 55 |
+
+Input totals include cache reads; the route's cache accounting is inconsistent
+between arms (see S5), so only totals are given.
+
+The two orientation rejections were stated omissions under the version 2
+rules: both answers said source is data, not instructions, without saying it
+grants no authority; the Context answer also never said a cursor is bound to
+its search term, and the Graphify answer did not say search re-inspects source.
+The reviewer passed the Context answer's freshness dimension while calling it
+borderline, so some reviewer discretion remains.
+
+## Tool friction seen in the Context arm
+
+The Context orientation session took 44 turns against plain's 29. Four of its
+failed or wasted calls came from the tools, not the model, and are fixed in the
+same change as this record:
+
+- Three `repository_packet` calls ran past the end of a file and the error did
+  not give the length; the model then searched for `the`, `export`, `async` and
+  `class` to learn the file's shape. The error now states the line count.
+- Two concurrent `repository_packet` calls were refused as already in progress.
+  Concurrent calls now wait in order.
+- `repository_search` rejected `local-source-unsigned`, the literal the task
+  asks about, because terms had to be one identifier. Literals whose ends fall
+  on whole tokens are now accepted.
+- `repository_coverage` rejected a draft naming more than eight symbols. It now
+  checks the first eight and lists the rest as not checked.
+
+The model also called `repository_explore` once against fifteen searches. That
+is left to the repeated runs to measure. Neither omission that failed the
+answer was a retrieval gap: the relevant source was in packets the model had
+fetched.

docs/experiments/smoke-flash-20260923/context-instructions.txt

@@ -0,0 +1,9 @@
+Use the configured z1p-repository tools for source discovery. First check that
+repository_status.root is this exact checkout or worktree; stop using a
+mismatched binding. Explore each symbol once, fetch the blocks you rely on with
+repository_packet and cite those lines. Run repository_coverage on the draft and
+fix what it reports. Read files directly only for evidence the tools cannot
+supply; tiny known-file edits need no scan. Refresh and re-fetch after edits,
+branch switches, pulls, merges or rebases; reconnect after changing the binding
+or server build. A shell directory change does not retarget Context. Keep the
+project's existing models, tests and review standards.

docs/experiments/smoke-flash-20260923/protocol.json

@@ -0,0 +1,171 @@
+{
+  "version": 1,
+  "experimentId": "three-way-deepseek-v4.1-flash-smoke-20260923",
+  "lockedAt": "2026-09-22T23:49:25.199Z",
+  "shellPathRule": "The client's Bash tool rebuilds PATH from the login profile, so every arm receives PATH through the settings env: the Node 24 bin directory plus the runner's PATH. The graphify arm additionally prepends the isolated virtualenv bin so `graphify` resolves. Verified by a probe before lock.",
+  "purpose": "Smoke test of the repeated-run configuration on a DeepSeek V4.1 Flash executor: checks that checker-and-scope code acceptance and version 2 structured rubrics run end to end in all three arms. Two tasks, one repetition; not a comparison and not reported as one.",
+  "taskPack": {
+    "directory": "../d5-20260921",
+    "reused": [
+      "tasks/*.json",
+      "setup/*.patch",
+      "acceptance/*.json",
+      "prepare-arm.mjs",
+      "accept.mjs"
+    ],
+    "note": "The eight locked v1 task definitions, seeded patches, frozen archives and deterministic checker are reused unchanged. v1, v2 and v3 results are not pooled with this experiment."
+  },
+  "tasks": [
+    "orientation-context",
+    "code-change-context"
+  ],
+  "retrievalInstructionRule": "Each arm's task prompt begins with one retrieval instruction of similar length and force, as the locked v1 measurement definition prescribed. Pilot 3 showed that the tools' own always-on text alone was ignored by the model in every arm, which measures nothing about the tools.",
+  "arms": {
+    "plain": {
+      "description": "Built-in Read, Edit, Write and Bash only. No MCP servers, no Graphify on PATH.",
+      "systemPromptAppendix": "Retrieval: use the built-in Read, Grep, Glob and Bash tools for source discovery.",
+      "retrievalInstruction": "you must use only bounded grep or rg and exact file reads for source discovery, confined to the task selection policy."
+    },
+    "graphify": {
+      "description": "Built-in tools plus a Graphify graph prebuilt in graphify-out/ with `graphify update <workspace>` (no model, no API key) and the graphify CLI on PATH. The system prompt appendix is Graphify's own always-on instruction text (always_on/claude-md.md) verbatim. graphify-out/ is hidden from Git through an environment-supplied core.excludesFile so no hashed file changes.",
+      "retrievalInstruction": "a Graphify knowledge graph is prebuilt in graphify-out/ and the graphify CLI is on PATH. Before any grep, find, cat or file read you must run `graphify query \"<question>\"` and `graphify explain \"<symbol>\"` (and `graphify path \"<A>\" \"<B>\"` for relationships) for the concepts in this task, then read only the file locations they return, confined to the task selection policy.",
+      "package": "graphifyy",
+      "packageVersion": "0.9.65",
+      "build": "graphify update <workspace>",
+      "instructionSource": "graphify/always_on/claude-md.md",
+      "instructionSha256": "5a599e9f148ebbf0a8786c4508c3954daf95d0d74b88223d890dbf37aad23ec6",
+      "graphOutExcludedFromGit": true
+    },
+    "context": {
+      "description": "Built-in tools plus the z1p-repository stdio MCP server (repository_status, repository_refresh, repository_explore, repository_coverage, repository_search, repository_packet) bound to the arm workspace, with the task selection policy in .z1p-navigation.json. Search and packet responses render as compact text; repository_explore answers one symbol per call. The system prompt appendix is the project instruction text from docs/GETTING-STARTED.md step 3 verbatim (context-instructions.txt). repository_coverage is a deterministic pre-submit check that lists explored files a draft answer leaves uncited.",
+      "package": "@forgesworn/context-tools",
+      "packageVersion": "source build at the recorded commit (repository_explore, repository_coverage with exact-quote check, compact rendering, pathPrefix); clean working tree, identified by dist digests below",
+      "server": "encrypted-context.mjs navigate <workspace>",
+      "instructionSource": "docs/GETTING-STARTED.md step 3 (context-instructions.txt in this directory)",
+      "retrievalInstruction": "the z1p-repository MCP tools are bound to this workspace. Before any grep, find, cat or file read you must call repository_status (and repository_refresh if it is unavailable, stale or unknown), then call repository_explore for each identifier in this task, use repository_search with pathPrefix only for literals, and obtain the source and tests you cite with repository_packet using the current expectedGeneration, reading files directly only for evidence those tools cannot supply, confined to the task selection policy. Before writing answer.json, call repository_coverage with your draft answer and the task's identifiers, then address each missing file or state why it does not bear on the task.",
+      "instructionSha256": "cb7f696ea84c42f5923cfc0470a5416b1b6053f4df29bb162e80bb8795b9f934",
+      "build": {
+        "headCommit": "7d436fc1c17fca78ac39bf74e74b9a87f33ed1a1",
+        "workingTreeClean": true,
+        "dist": {
+          "repository-coverage.js": "3a112ab4f27478836a9a5c785a4875d40922d4e4b6ce1f7e1bc499470a72df75",
+          "repository-explore.js": "54269956eee2297c8318c5fd430c2bdd0417ac16298027b52ce8840fbb9e3249",
+          "repository-navigation.js": "79459fa72f754522fdd4b2858c82fde73298d1480bb6af1bec37b341b69daafb",
+          "repository-navigation-mcp.js": "ad57575b2b1cac2c75a94491f0b16d40467564dd22c5a7ebf1ea8b2ffbec442c"
+        }
+      }
+    }
+  },
+  "armOrders": [
+    [
+      "plain",
+      "graphify",
+      "context"
+    ],
+    [
+      "graphify",
+      "context",
+      "plain"
+    ],
+    [
+      "context",
+      "plain",
+      "graphify"
+    ]
+  ],
+  "armOrderRule": "Task i uses armOrders[i mod 3]. Arms of one task run sequentially in that order; each arm has its own prepared workspace, fresh headless session and, for the context arm, its own server process. No answer, diff, index or transcript is shared between arms.",
+  "executor": {
+    "client": "claude-code",
+    "clientVersion": "2.1.280",
+    "model": "deepseek-v4.1-flash:cloud",
+    "effort": "medium",
+    "maxTurns": 80,
+    "maxBudgetUsd": null,
+    "settingSources": "project",
+    "settingsOverride": {
+      "enabledPlugins": {
+        "agents-md@builtin": false
+      }
+    },
+    "settingsOverrideReason": "A probe showed the built-in agents-md plugin injects the frozen tree's AGENTS.md into every arm; the Context revision's AGENTS.md instructs use of z1p-repository tools, so it is disabled for all arms. A second probe confirmed no project instruction text remains in context.",
+    "disallowedTools": [
+      "WebFetch",
+      "WebSearch",
+      "Agent",
+      "ToolSearch",
+      "Workflow",
+      "EnterWorktree",
+      "ExitWorktree",
+      "RemoteTrigger",
+      "CronCreate",
+      "CronDelete",
+      "CronList",
+      "ScheduleWakeup",
+      "SendMessage",
+      "PushNotification",
+      "Monitor",
+      "DesignSync",
+      "ListAgents",
+      "TaskStop",
+      "NotebookEdit"
+    ],
+    "builtInToolsObserved": "This Claude Code build exposes Bash, Read, Edit and Write for source work; Grep and Glob are not separate tools, so plain-arm search runs through Bash.",
+    "permissions": "dangerously-skip-permissions inside a disposable frozen workspace",
+    "skills": "disabled",
+    "mcp": "strict; only the arm's --mcp-config",
+    "sessionPersistence": false,
+    "promptDelivery": "identical task prompt on stdin for every arm; the only difference is the arm's system prompt appendix, PATH and MCP config",
+    "modelRoute": "Claude Code 2.1.280 with ANTHROPIC_BASE_URL pointing at the local Ollama 0.34.2 daemon (Anthropic-compatible /v1/messages), which proxies the cloud tag to ollama.com; verified by a smoke test with Bash and MCP tool calls",
+    "env": {
+      "ANTHROPIC_BASE_URL": "http://127.0.0.1:11435",
+      "ANTHROPIC_API_KEY": "ollama",
+      "ANTHROPIC_AUTH_TOKEN": "ollama"
+    },
+    "costNote": "The client reports a list-price estimate for an unrecognised model; it is meaningless here and no budget cap is applied. Provider usage counters (input, cache read, output) are recorded; Ollama Cloud credit consumption is read from the account, not from receipts.",
+    "effortNote": "--effort medium is accepted by the client; whether the route honours it is unknown. The smoke test reported zero thinking tokens."
+  },
+  "reviewer": {
+    "client": "claude-code",
+    "model": "claude-sonnet-5",
+    "effort": "high",
+    "maxTurns": 3,
+    "disallowedTools": [
+      "Bash",
+      "Read",
+      "Edit",
+      "Write",
+      "Grep",
+      "Glob",
+      "TodoWrite"
+    ],
+    "retry": "one further attempt if the structured verdict does not parse; both attempts are recorded and their usage and time are summed",
+    "tools": "none (Bash, Read, Edit, Write, Grep and Glob disallowed as well); receives the answer or diff, the private rubric, the deterministic checker result and bounded excerpts of the cited frozen source",
+    "blind": "does not see the arm name or any other arm's output"
+  },
+  "measures": {
+    "accepted": "code tasks: deterministic checker and scope check passed (no model reviewer); structured tasks: checker passed and the blind reviewer accepted against rubric version 2 with no material issue",
+    "executorToolCalls": "count of tool_use blocks by tool name from the stream",
+    "executorInputTokens": "input_tokens + cache_creation_input_tokens + cache_read_input_tokens from the result usage; uncached = input_tokens + cache_creation_input_tokens",
+    "executorOutputTokens": "output_tokens from the result usage",
+    "costEstimateUsd": "Claude Code total_cost_usd, a list-price estimate, not subscription billing",
+    "executorSeconds": "wall time of the headless session",
+    "armSeconds": "graphify build (arm-specific setup) + executor + checker + reviewer wall time",
+    "reviewerTokens": "reviewer session usage",
+    "setupSeconds": "dependency installation, shared by all arms and excluded"
+  },
+  "decisionRule": "None: a smoke test. It passes if every arm completes, every code cell records a scope result and no reviewer, and every structured cell has a parsed verdict whose prompt contains the version 2 review rules.",
+  "knownAsymmetries": [
+    "Graphify indexes the whole frozen workspace; Context indexes only the task selection policy. Graphify is not disadvantaged by this.",
+    "Graphify's arm-specific build time is counted in armSeconds; Context's refresh happens inside the session and is counted as a tool call.",
+    "Arm instructions are delivered by system prompt appendix rather than a CLAUDE.md file so the frozen tree stays identical across arms."
+  ],
+  "derivedFrom": "three-way-deepseek-v4-pro-20260922-s5",
+  "codeAcceptance": "checker-and-scope",
+  "rubricDir": "../rubric-v2-20260923",
+  "changesFromS5": [
+    "Executor deepseek-v4.1-flash:cloud (same route and effort).",
+    "Two tasks, one repetition.",
+    "Code acceptance by checker plus scope check; structured rubrics version 2.",
+    "Context arm at the current build with repository_coverage; instruction text is docs/GETTING-STARTED.md step 3 as of the recorded commit; retrieval instruction as v3."
+  ]
+}

packages/context-tools/README.md

@@ -65,7 +65,8 @@ over an unsigned in-memory index of that root: `repository_status`,
 references with enclosing declarations, importing files and tests),
 `repository_coverage` (which explored files a draft answer leaves uncited,
 and whether each cited token is quoted exactly),
-`repository_search` (exact identifier lines grouped by file, with `pathPrefix`)
+`repository_search` (exact identifier or literal lines grouped by file, with
+`pathPrefix`)
 and `repository_packet` (verbatim ranges or complete TypeScript/JavaScript
 blocks). Search, explore and packet responses are compact text by default and
 JSON with `format: "json"`. See the