fix: resolve 14 test failures, add coverage tooling, and align docs with codebase#35
Conversation
…ith codebase - Add ShadowContextPacket type and buildUserMessage function to prompt-builder.ts (fixes missing exports that caused 5 test failures across prompt-builder and inference tests) - Add packContext wrapper to context-packager.ts with richer return shape (fixes 7 test failures in inference-contract tests) - Refactor buildContextPacket to delegate to packContext (eliminates code duplication) - Re-export SHADOW_SYSTEM_PROMPT from prompt-builder.ts for import convenience - Update shadow-mcp-server.ts to import from prompt-builder instead of prompts - Fix inference-contract test expectations to match actual tool history behavior - Add test:coverage script with @vitest/coverage-v8 dependency - Add npm test step to CI workflow (prompt-parity.yml) - Add npm test to pre-commit hook - Add coverage/ to .gitignore - Update docs: architecture.md, domain-inference.md, getting-started.md, todo.md to reflect current codebase state (merged PRs, actual file layout, Phase 1 completion) - Clarify third_party/ usage policy in AGENTS.md All 145 tests pass. Build succeeds. Prompts in sync. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
|
Warning Rate limit exceeded
Your organization is not enrolled in usage-based pricing. Contact your admin to enable usage-based pricing to continue reviews beyond the rate limit, or try again in 52 minutes and 23 seconds. ⌛ How to resolve this issue?After the wait time has elapsed, a review can be triggered using the We recommend that you space out your commits to avoid hitting the rate limit. 🚦 How do rate limits work?CodeRabbit enforces hourly rate limits for each developer per organization. Our paid plans have higher rate limits than the trial, open-source and free plans. In all cases, we re-allow further reviews after a brief timeout. Please see our FAQ for further information. ℹ️ Review info⚙️ Run configurationConfiguration used: Organization UI Review profile: ASSERTIVE Plan: Pro Run ID: 📒 Files selected for processing (2)
📝 WalkthroughWalkthroughThis PR refactors the inference module's architecture by establishing Changes
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes Possibly related issues
Possibly related PRs
Poem
🚥 Pre-merge checks | ✅ 2 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: fb014935be
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| @@ -1,2 +1,3 @@ | |||
| #!/usr/bin/env sh | |||
| npm run prompts:check | |||
| npm test --prefix shadow-agent | |||
There was a problem hiding this comment.
Because this hook script does not enable set -e (and commands are on separate lines), a failing npm run prompts:check is ignored if npm test --prefix shadow-agent succeeds, so commits can pass with prompt artifacts out of sync. This was introduced when the second command was added; chain commands with && or enable fail-fast behavior so any failed check blocks the commit.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Pull request overview
This PR repairs broken test imports/exports in the inference stack, adds coverage tooling + CI/pre-commit test execution, and updates documentation to match what’s actually shipped in main.
Changes:
- Fixes inference test failures by moving the canonical
ShadowContextPacket/buildUserMessageexports intoprompt-builder.ts, updating imports, and addingpackContext()tocontext-packager.ts. - Adds
vitestcoverage support plus CI and pre-commit hooks to runnpm test(and build in CI). - Aligns workspace/app/docs READMEs and inference/architecture docs with the current implementation status.
Reviewed changes
Copilot reviewed 16 out of 17 changed files in this pull request and generated 4 comments.
Show a summary per file
| File | Description |
|---|---|
| shadow-agent/tests/inference/inference-contract.test.ts | Updates imports and tool-history expectations to match current pairing behavior. |
| shadow-agent/tests/inference.test.ts | Consolidates prompt imports via prompt-builder. |
| shadow-agent/src/mcp/shadow-mcp-server.ts | Switches SHADOW_SYSTEM_PROMPT import to prompt-builder. |
| shadow-agent/src/inference/prompt-builder.ts | Introduces canonical context packet type and user-message rendering (with privacy gating). |
| shadow-agent/src/inference/context-packager.ts | Adds packContext() API and refactors buildContextPacket() to delegate to it. |
| shadow-agent/package.json | Adds test:coverage script and coverage dependency. |
| shadow-agent/package-lock.json | Locks @vitest/coverage-v8 and transitive deps. |
| shadow-agent/README.md | Clarifies relationship between workspace README and app README. |
| shadow-agent/.gitignore | Ignores coverage/ output directory. |
| docs/todo.md | Updates project status tracking and adds infra notes for this remediation. |
| docs/getting-started.md | Corrects project structure to match current main. |
| docs/domain-inference.md | Updates inference file map/status to reflect implemented modules. |
| docs/architecture.md | Updates phase status and roadmap bullets to reflect what’s merged vs feature branches. |
| README.md | Clarifies workspace reading order and third_party/ expectations. |
| AGENTS.md | Clarifies third_party/ as reference-only (no runtime imports). |
| .husky/pre-commit | Adds shadow-agent test run to pre-commit. |
| .github/workflows/prompt-parity.yml | Expands prompt-parity into CI running prompt checks + tests + build. |
Files not reviewed (1)
- shadow-agent/package-lock.json: Language not supported
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| const delivery = options.delivery ?? 'local'; | ||
| const privacy = options.privacy ?? { allowRawTranscriptStorage: false, allowOffHostInference: false }; | ||
|
|
||
| if (delivery === 'off-host' && !privacy.allowOffHostInference) { | ||
| throw new Error('Off-host inference is disabled until the user explicitly opts in.'); | ||
| } | ||
|
|
||
| const sanitize = (text: string): string => { | ||
| if (delivery === 'off-host' && privacy.allowOffHostInference && options.includeRawTranscript && privacy.allowRawTranscriptStorage) { | ||
| return text; | ||
| } | ||
| return sanitizeTranscriptText(text); | ||
| }; |
There was a problem hiding this comment.
includeRawTranscript is never validated against privacy.allowRawTranscriptStorage. If a caller requests raw transcript inclusion off-host but hasn’t opted into raw transcript handling, this currently silently falls back to sanitization instead of failing fast. Consider reusing assertOffHostInferenceAllowed(privacy, { includeRawTranscript }) (and DEFAULT_TRANSCRIPT_PRIVACY_SETTINGS) or adding an explicit check so behavior matches the privacy module’s contract.
| '', | ||
| `--- Recent Events (${packet.recentEvents.length}) ---`, | ||
| ...packet.recentEvents.map((e) => | ||
| `${e.timestamp} [${e.kind}] ${e.actor}: ${sanitize(JSON.stringify(e.payload).slice(0, 120))}` |
There was a problem hiding this comment.
The payload preview is truncated before sanitization (JSON.stringify(e.payload).slice(0, 120)), which can leak partial secrets (e.g., a token prefix that no longer matches the redaction regex). Sanitize first, then truncate (or sanitize the structured payload via sanitizeCanonicalEvent / sanitizeUnknown before stringifying).
| `${e.timestamp} [${e.kind}] ${e.actor}: ${sanitize(JSON.stringify(e.payload).slice(0, 120))}` | |
| `${e.timestamp} [${e.kind}] ${e.actor}: ${sanitize(JSON.stringify(e.payload)).slice(0, 120)}` |
| while (approximateTokens > tokenBudget && trimOldest()) { | ||
| approximateTokens = estimateTokens(truncatedPacket); | ||
| } | ||
|
|
||
| return truncatedPacket; | ||
| return { packet: truncatedPacket, truncated: true, approximateTokens }; |
There was a problem hiding this comment.
packContext can return approximateTokens still above tokenBudget when trimOldest() can no longer remove items (e.g., very small tokenBudget values). That breaks the implied contract of “respecting tokenBudget”. Consider detecting this case after the loop and either (a) further truncating strings, (b) returning a minimal packet, or (c) returning an additional flag/error indicating the budget cannot be met.
| @@ -1,2 +1,3 @@ | |||
| #!/usr/bin/env sh | |||
There was a problem hiding this comment.
This hook now runs multiple commands, but the script doesn’t set -e or chain with &&. If npm run prompts:check fails but npm test passes, the hook may exit 0 (since the last command succeeds), allowing commits that break prompt parity. Add set -e at the top (or chain commands with &&) so any failure blocks the commit.
| #!/usr/bin/env sh | |
| #!/usr/bin/env sh | |
| set -e |
There was a problem hiding this comment.
Actionable comments posted: 3
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In @.husky/pre-commit:
- Around line 1-3: Add "set -e" immediately after the shebang so the pre-commit
hook exits on the first failing command; update the pre-commit script that
currently runs npm run prompts:check followed by npm test --prefix shadow-agent
to include set -e (so a failing npm run prompts:check cannot be masked by the
later npm test command).
In `@shadow-agent/src/inference/prompt-builder.ts`:
- Around line 39-40: The default privacy object is being re-declared inline in
prompt-builder.ts; instead import and use the canonical
DEFAULT_TRANSCRIPT_PRIVACY_SETTINGS exported from
shadow-agent/src/shared/privacy.ts and use it in the privacy fallback (replace
the literal used in the privacy const). Update imports to include
DEFAULT_TRANSCRIPT_PRIVACY_SETTINGS and change the privacy assignment to default
to that constant so future fields won't drift (reference symbols: privacy const
in prompt-builder.ts and DEFAULT_TRANSCRIPT_PRIVACY_SETTINGS from
shared/privacy).
- Around line 46-51: The sanitize() predicate currently gates raw-text
passthrough on delivery === 'off-host', which prevents local raw transcripts
even when privacy.allowRawTranscriptStorage and options.includeRawTranscript are
true; change the condition in the sanitize function to only check the privacy
flags (options.includeRawTranscript && privacy.allowRawTranscriptStorage) so raw
content is allowed independent of delivery mode. Replace the duplicated
DEFAULT_TRANSCRIPT_PRIVACY_SETTINGS with an import from ../shared/privacy and
use that constant where defaults are applied. Also rename the options flag
includeRawTranscript to a clearer name (e.g., includeRawPayloads or
includeRawEventPayloads) across prompt-builder.ts and related callers to reflect
that it controls event payloads, tool args, and transcript text, updating
references and types accordingly.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: ASSERTIVE
Plan: Pro
Run ID: 881974fd-afbc-400a-89cc-d6c3242ee83b
⛔ Files ignored due to path filters (1)
shadow-agent/package-lock.jsonis excluded by!**/package-lock.json
📒 Files selected for processing (16)
.github/workflows/prompt-parity.yml.husky/pre-commitAGENTS.mdREADME.mddocs/architecture.mddocs/domain-inference.mddocs/getting-started.mddocs/todo.mdshadow-agent/.gitignoreshadow-agent/README.mdshadow-agent/package.jsonshadow-agent/src/inference/context-packager.tsshadow-agent/src/inference/prompt-builder.tsshadow-agent/src/mcp/shadow-mcp-server.tsshadow-agent/tests/inference.test.tsshadow-agent/tests/inference/inference-contract.test.ts
| const sanitize = (text: string): string => { | ||
| if (delivery === 'off-host' && privacy.allowOffHostInference && options.includeRawTranscript && privacy.allowRawTranscriptStorage) { | ||
| return text; | ||
| } | ||
| return sanitizeTranscriptText(text); | ||
| }; |
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
# First, locate and read the prompt-builder.ts file around the flagged lines
find . -name "prompt-builder.ts" -type fRepository: Coldaine/AgentVisualCrazy
Length of output: 115
🏁 Script executed:
# Also check for documentation files mentioned
find . -name "domain-inference.md" -o -name "architecture.md" | head -20Repository: Coldaine/AgentVisualCrazy
Length of output: 118
🏁 Script executed:
# Search for DEFAULT_TRANSCRIPT_PRIVACY_SETTINGS to understand the duplication claim
rg "DEFAULT_TRANSCRIPT_PRIVACY_SETTINGS" --type ts --type tsx -A 3 -B 1Repository: Coldaine/AgentVisualCrazy
Length of output: 96
🏁 Script executed:
# Corrected search for DEFAULT_TRANSCRIPT_PRIVACY_SETTINGS
rg "DEFAULT_TRANSCRIPT_PRIVACY_SETTINGS" --type ts -A 3 -B 1Repository: Coldaine/AgentVisualCrazy
Length of output: 3365
🏁 Script executed:
# Read prompt-builder.ts - check file size first
wc -l ./shadow-agent/src/inference/prompt-builder.tsRepository: Coldaine/AgentVisualCrazy
Length of output: 118
🏁 Script executed:
# Read the critical section and surrounding context (lines 35-91)
sed -n '35,91p' ./shadow-agent/src/inference/prompt-builder.tsRepository: Coldaine/AgentVisualCrazy
Length of output: 1973
🏁 Script executed:
# Also check the imports at the top
head -30 ./shadow-agent/src/inference/prompt-builder.tsRepository: Coldaine/AgentVisualCrazy
Length of output: 1264
🏁 Script executed:
# Read documentation files to verify privacy matrix claims
cat ./docs/architecture.mdRepository: Coldaine/AgentVisualCrazy
Length of output: 8138
🏁 Script executed:
cat ./docs/domain-inference.mdRepository: Coldaine/AgentVisualCrazy
Length of output: 5107
🏁 Script executed:
# Search for BuildUserMessageOptions interface to see full signature
rg "BuildUserMessageOptions" shadow-agent/src/inference/prompt-builder.ts -A 10Repository: Coldaine/AgentVisualCrazy
Length of output: 839
Clarify privacy matrix: raw transcript availability is gated to off-host delivery, but docs describe it as a separate opt-in.
The sanitize condition at line 47 requires delivery === 'off-host', which means local prompts always sanitize even when allowRawTranscriptStorage: true and includeRawTranscript: true. However, architecture.md and domain-inference.md both describe "raw transcript storage/export" as a separate explicit opt-in, which could reasonably be interpreted as independent from transport mode. If local-raw should be allowed, relax the condition to gate only on privacy flags:
if (options.includeRawTranscript && privacy.allowRawTranscriptStorage) {
Also, line 40 duplicates DEFAULT_TRANSCRIPT_PRIVACY_SETTINGS from ../shared/privacy — import and use the constant instead to keep defaults aligned. Finally, the option name includeRawTranscript understates scope: it controls sanitization of event payloads, tool arguments, and transcript text.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@shadow-agent/src/inference/prompt-builder.ts` around lines 46 - 51, The
sanitize() predicate currently gates raw-text passthrough on delivery ===
'off-host', which prevents local raw transcripts even when
privacy.allowRawTranscriptStorage and options.includeRawTranscript are true;
change the condition in the sanitize function to only check the privacy flags
(options.includeRawTranscript && privacy.allowRawTranscriptStorage) so raw
content is allowed independent of delivery mode. Replace the duplicated
DEFAULT_TRANSCRIPT_PRIVACY_SETTINGS with an import from ../shared/privacy and
use that constant where defaults are applied. Also rename the options flag
includeRawTranscript to a clearer name (e.g., includeRawPayloads or
includeRawEventPayloads) across prompt-builder.ts and related callers to reflect
that it controls event payloads, tool args, and transcript text, updating
references and types accordingly.
- Add set -e to pre-commit hook so prompts:check failure blocks commit - Sanitize event payload before truncating to prevent partial secret leakage - Import DEFAULT_TRANSCRIPT_PRIVACY_SETTINGS from shared/privacy instead of re-declaring the default literal inline Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Nine drift items were identified during the 2026-04-19 end-of-day review (captured in docs/third-party-readiness-briefing.html, added here). Each is addressed as a targeted edit rather than a rewrite -- the docs are mostly right, they were just out of step with what PRs #28/#29/#31/#32/#34/#35 actually landed. Changes: - architecture.md: "Phase 2 (Release Candidate)" was misleading -- Canvas2D (#26) and live capture (#27) are still on branches. Relabel to "Phase 2 (In Progress)" and state the two PRs that close the phase. Also replaces the Three.js-only background claim with the Citadel-dot-grid-preferred decision from plan-gui-rendering.md, and converts the four section-sign cross-refs to real Markdown links. - domain-gui.md / domain-events.md / domain-inference.md: unify status headers under one schema (Planned / Partial on main / Landed on main) instead of the three different styles previously in use. - history/log.md: fix ascending chronological ordering (04-18 was before 04-12) and add the missing 2026-04-19 entry covering PRs #29, #31, #32, #34, #35 merged that day. - plan-master-a-must-do.md: RepoVis archival, temp-file removal, and domain status headers were listed as TODO but were all completed 2026-04-18/19. Mark them done with audit-trail strikethrough. Also refresh the Pre-Commit/CI section (husky -> .githooks migration) and the Definition of Done (five PRs merged, four remain). - plan-master-coordination.md: remove the three empty Stream A/B/C scratchpad placeholders and replace the Round 2 Dispatch TODO list with a closeout record pointing to the promoted outputs. - plan-gui-rendering.md / plan-event-capture.md / plan-inference-engine.md: convert the remaining section-sign cross-refs into Markdown links with descriptive section names. - getting-started.md: cd into shadow-agent for test commands, note the PR #35 bisect caveat, and mention the CI/pre-commit gates. - prompts/shadow-system-prompt.json: append a 2026-04-19 iteration-log entry documenting the regeneration; regenerated docs/prompts/ shadow-system-prompt.md is in sync (npm run prompts:check passes). - docs/third-party-readiness-briefing.html: commit the HTML artifact that drove this pass so the reasoning behind these edits is auditable. Validation: 145 tests pass (npm test), npm run prompts:check clean, no lint errors. Scope note: open PRs #26, #27, #30, #33 are deliberately untouched -- those are thousand-plus-line product-feature or infrastructure branches and each deserves its own focused session. Made-with: Cursor
* docs: remediate drift between documentation and current main state Nine drift items were identified during the 2026-04-19 end-of-day review (captured in docs/third-party-readiness-briefing.html, added here). Each is addressed as a targeted edit rather than a rewrite -- the docs are mostly right, they were just out of step with what PRs #28/#29/#31/#32/#34/#35 actually landed. Changes: - architecture.md: "Phase 2 (Release Candidate)" was misleading -- Canvas2D (#26) and live capture (#27) are still on branches. Relabel to "Phase 2 (In Progress)" and state the two PRs that close the phase. Also replaces the Three.js-only background claim with the Citadel-dot-grid-preferred decision from plan-gui-rendering.md, and converts the four section-sign cross-refs to real Markdown links. - domain-gui.md / domain-events.md / domain-inference.md: unify status headers under one schema (Planned / Partial on main / Landed on main) instead of the three different styles previously in use. - history/log.md: fix ascending chronological ordering (04-18 was before 04-12) and add the missing 2026-04-19 entry covering PRs #29, #31, #32, #34, #35 merged that day. - plan-master-a-must-do.md: RepoVis archival, temp-file removal, and domain status headers were listed as TODO but were all completed 2026-04-18/19. Mark them done with audit-trail strikethrough. Also refresh the Pre-Commit/CI section (husky -> .githooks migration) and the Definition of Done (five PRs merged, four remain). - plan-master-coordination.md: remove the three empty Stream A/B/C scratchpad placeholders and replace the Round 2 Dispatch TODO list with a closeout record pointing to the promoted outputs. - plan-gui-rendering.md / plan-event-capture.md / plan-inference-engine.md: convert the remaining section-sign cross-refs into Markdown links with descriptive section names. - getting-started.md: cd into shadow-agent for test commands, note the PR #35 bisect caveat, and mention the CI/pre-commit gates. - prompts/shadow-system-prompt.json: append a 2026-04-19 iteration-log entry documenting the regeneration; regenerated docs/prompts/ shadow-system-prompt.md is in sync (npm run prompts:check passes). - docs/third-party-readiness-briefing.html: commit the HTML artifact that drove this pass so the reasoning behind these edits is auditable. Validation: 145 tests pass (npm test), npm run prompts:check clean, no lint errors. Scope note: open PRs #26, #27, #30, #33 are deliberately untouched -- those are thousand-plus-line product-feature or infrastructure branches and each deserves its own focused session. Made-with: Cursor * docs: address Copilot + Codex review feedback on drift-remediation PR Fixes all six inline comments from the automated reviewers on PR #36. Root cause for four of them was my conflation of the in-flight husky -> .githooks migration (which is stashed, not on this branch) with current reality on main. The other two were residual section-sign cross-refs that my first sweep missed. - getting-started.md: reference the actual `.husky/pre-commit` hook (not a speculative `.githooks/pre-commit`); name the CI workflow correctly (it is `CI`, defined in `.github/workflows/prompt-parity.yml`); accurately describe what the hook runs (`prompts:check` and `npm test --prefix shadow-agent`). - plan-master-a-must-do.md: drop the incorrect "hook has since migrated to .githooks/pre-commit" claim; describe the current `.husky/pre-commit` instead. Convert the remaining "under section 3" cross-ref to a clickable anchor link. - plan-master-coordination.md: replace the Stream A/B/C "section 1 / 2 / 3" cross-refs with descriptive Markdown links to the actual files and named sections. Same fix applied to the Round 2 Dispatch PR-fix bullet. - architecture.md: replace the "section 8" cross-ref with the full step heading "Optional Background Atmosphere Layer (Choose One)" as a descriptive link target. Validation: rg for section-sign notation returns zero hits; `npm run prompts:check` clean; `npm test` 145/145 pass. Made-with: Cursor
Summary
Fixes all 14 test failures on
main, adds missing test infrastructure, and aligns documentation with the current codebase state.Test Fixes (14 failures → 0)
Missing exports from
prompts.ts(5 failures)ShadowContextPackettype andbuildUserMessagefunction toprompt-builder.ts(notprompts.ts, since that file is generated and shouldn't contain hand-written code)SHADOW_SYSTEM_PROMPTfromprompt-builder.tsfor import convenienceshadow-mcp-server.tsto import fromprompt-builderinstead ofpromptsMissing
packContextincontext-packager.ts(7 failures)packContextfunction with richer return shape:{ packet, truncated, approximateTokens }buildContextPacketto delegate topackContext(eliminates ~90 lines of duplication)PackContextOptionsandPackContextResultinterfacesTest expectation mismatches (2 failures)
inference-contract.test.tstool history expectations to match actualtool_started→tool_completed/tool_failedpairing behaviorInfrastructure
test:coveragescript: Added topackage.jsonwith@vitest/coverage-v8dependency (61% statement coverage)npm teststep toprompt-parity.ymlworkflow (previously only checked prompt sync)npm testin addition toprompts:check.gitignore: Addedcoverage/directoryDocumentation Alignment
architecture.md: Updated Phase 1 to "Completed", Phase 2 status to reflect inference scaffolding on maindomain-inference.md: Updated file map to match actual exports, removed non-existentopencode-client.tsgetting-started.md: Corrected project structure description (Canvas2D renderer not yet on main)todo.md: Marked merged PRs as completed, updated status trackingAGENTS.md: Clarifiedthird_party/is reference material, not runtime dependencyVerification
Co-authored-by: Copilot 223556219+Copilot@users.noreply.github.com
PR Summary: Fix 14 Test Failures, Add Coverage Tooling, and Align Docs with Codebase
Overview
This PR resolves 14 failing tests, implements missing inference infrastructure exports, adds coverage tooling and CI integration, and updates documentation to reflect the current codebase state. All 145 tests now pass.
Key Changes by Category
🔧 Core Inference Module Updates
SHADOW_SYSTEM_PROMPTfrom./prompts• Added
ShadowContextPackettype (canonical inference context shape)• Implemented
buildUserMessage(packet, options)with privacy controls• Added
BuildUserMessageOptionsinterface•
buildInferenceRequest()now uses localbuildUserMessageimplementationShadowContextPacketimport source toprompt-builder• Introduced
packContext(state, events, options)returning{ packet, truncated, approximateTokens }• Refactored
buildContextPacket()to delegate topackContext().packet• Added
PackContextOptions(configurabletokenBudget,recentWindowSize)• Added
PackContextResultinterface• Dynamic truncation logic based on
approximateTokensvstokenBudgetSHADOW_SYSTEM_PROMPTimport source frompromptstoprompt-builder🧪 Test Infrastructure
test:coveragescript:vitest run --coverage• Added
@vitest/coverage-v8dev dependencySHADOW_SYSTEM_PROMPTimport source toprompt-builderSHADOW_SYSTEM_PROMPT,buildUserMessage,ShadowContextPacketfromprompt-builder• Adjusted "builds tool history" test: added
tool_startedevent for bash; updated expectations from 3-entry to 2-entrytoolHistorycoverage/directory🚀 CI/CD & Pre-commit Enhancements
• Renamed job from
prompt-paritytotest• Added
npm teststep• Added
npm run buildstep inshadow-agentdirectorynpm test --prefix shadow-agentto commit validation📚 Documentation Updates
• Clarified root README covers workspace layout;
shadow-agent/README.mdcovers app scope• Clarified
third_party/as disposable reference material (not runtime imports)• Updated description: patterns for inspection/porting only; can be removed once absorbed
third_party/is reference material, not product surface• Noted patterns should not be used for runtime imports
• Noted Canvas2D renderer and capture pipeline remain on feature branches
• Expanded Phase 2 inference-engine details: auth, context packaging, prompt building, response parsing, trigger logic, orchestrator, direct Anthropic fallback, MCP server
• Noted OpenCode client implementation incomplete
• Marked modules as implemented: auth loader, context-packager, prompt-builder, response parser, trigger logic, orchestrator, MCP server
• Explicitly marked
opencode-client.tsas not yet implementedshadow-agent/src/electron/responsibility: file loading (not filesystem watching)• Clarified Canvas2D visualization on feature branch; main uses simplified panels
• Expanded
shadow-agent/src/shared/scope: privacy, transcript parsing, replay store• Added
shadow-agent/src/inference/andsrc/mcp/sections• Marked several test tasks completed
• Updated OpenCode client status to "Not yet implemented"
• Marked Phase 2 capture/visual tests as "Blocked"
• Added Infrastructure section with coverage tooling, CI, and pre-commit hook improvements
Test Results
✅ 145/145 tests pass (was 14 failures)
Build Verification
✅ Vite renderer + esbuild Electron build succeeds
✅ Prompt artifacts in sync
✅ Coverage tooling functional (~61% statement coverage)