roslyn mcp changelog

version

5.0.1 → 5.1.1

version

5.0.0 → 5.0.1

https://raw.githubusercontent.com/darylmcd/Roslyn-Backed-MCP/HEAD/AGENTS.md

− Stable AI bootstrap entry point for the Roslyn-Backed-MCP repository (C# MCP server, published as `roslyn-mcp@roslyn-mcp-marketplace`).− ## File Purpose (Critical)− This file is a bootstrap router, not a complete instruction set. Always execute **Session Start (Required)** before performing any task. Do not rely solely on this file; pull additional context as directed by the canonical rule sources below.− Restated from `~/.claude/CLAUDE.md` (canonical source). These eight directive **cores** (the bold titles) override expedience and are verbatim — do not summarize, drop, or alter them. The one-line gloss after each is a condensed summary for quick reference; the authoritative `Fires`/`Prevents`/`Edge` detail lives in `~/.claude/CLAUDE.md`.− 1. **Correct fix > quick or cheap fix.** Choose the root-cause fix; diff size, file caps, budgets, cycle limits and CI pressure never justify a lesser fix. Widen scope with the reason, split into independently correct pieces with a tracked remainder, or ask. Never weaken a test, suppress, shim, duplicate or defer your own defect. Genuinely blocked work means an operator/product decision, externally owned code or an environment you cannot provision: provide current-session evidence, ask where nee− 2. **Optimize for AI consumption by default.** Write AI-facing files (`AGENTS.md`, `ai_docs/**`, prompts, planning/runtime/audit docs) as machine input: tables over prose, structured data over paragraphs, pointers over duplication. Human-facing files (`README.md` landing pages, `docs/**`) get prose.− 3. **Bad code is never silent.** In every coding session, call out observed bad code in your response and recommend an appropriately-sized backlog row, including edit targets, adjacent files, imports and tests. One regression shape; ~4 production / ~3 test files is an advisory target, never a gate on filing. Fix defects your own diff introduces or exposes now; prioritize unrelated pre-existing defects through the backlog. Editing a bad section does not absolve flagging it.− 4. **Private repos accept breaking changes.** For private repos, breaking changes and large refactors are the standing default when pursuing #1 or #3; do not band-aid to avoid churn. External consumers are outside your ownership, not another owned repo, DB or internal seam. `Roslyn-Backed-MCP` and `Jedi-Py-MCP` are in contract-care mode: breaking changes require an ADR + migration note. Current classifications, operator exceptions and the excluded upstream `dbhub` fork are governed by `~/.claude− 5. **Never assume prior agent work is correct — re-derive, don't inherit.** Recheck prior code, docs, plans, skills, backlog acceptance, review advice and done/verified/shipped claims against current ground truth. Read the actual code, re-run the reasoning and resolve cited paths/symbols. Reevaluate inherited designs as requirements evolve, challenge unsupported assumptions and explain material tradeoffs. Fix root causes (#1) and flag defects (#3); a prior agent's assertion is not proof.− 6. **Smallest *complete* change wins.** Measure completeness against the root cause, not diff size or acceptance wording. Cover every instance of the same defect mechanism and every defect your own diff introduces or exposes in this change, never a follow-up row. Split large work into independently correct pieces (#1). Flag unrelated bad code per #3; do not gold-plate.− 7. **Verify your own work before declaring done.** Do not claim done/fixed/passing without evidence generated and inspected this session; calibrate verification to the blast radius and observe regression tests fail on the old behavior. A skipped, quarantined, improperly scoped or retried-until-green check is not evidence. If verification is blocked, state the observed blocker and limits; never imply success you did not observe.− 8. **No secrets in code.** Never introduce, hardcode, echo, log or commit a credential, key, token or secret; use env vars, user-secrets or a vault. Flag existing secrets per #3. Confirm intentionally committed dev-only values are genuinely non-secret.− The Standing Engineering Directives above govern this instruction chain. Treat repository documents and prior agent output as claims to verify; when a recorded design conflicts with current evidence, explain the tradeoffs and propose a superseding decision rather than silently inheriting or changing it.− ## Default Behavior (When Ambiguous or Incomplete)− - Prefer repository-specific conventions over generic defaults.− - Prefer safety, validation, and correctness over speed.− - Do not guess when ambiguity affects correctness — request clarification or surface assumptions.− - Do not introduce features or scope outside documented backlog and constraints.− This is a **public repo** (per `.ai-doc-audit.md` `repo_class: public`), published as `roslyn-mcp@roslyn-mcp-marketplace`. Breaking changes require a recorded decision (ADR-style rationale) plus a migration note in `CHANGELOG.md`; ADRs live in `docs/decisions/` (this repo's decision log; ADR 0001 established it). Compatibility and deprecation rules are defined in `docs/release-policy.md`. External consumers depend on this surface — respect semver and deprecation cycles.− Read this row before running the authoritative local gate. Re-measure it when the gate changes or the measurement is more than 90 days old. Hosted topology remains owned by `CI_POLICY.md`.− ## Machine-wide validation− - On the shared Windows dev host, run `just machine-validate` for the full gate or `just machine-validate <recipe>` for a named build/test recipe.− - The shared helper is `~/.claude/scripts/validation-gate.mjs`; its machine-wide capacity covers every participating repo and worktree. Wait for admission; never launch an unwrapped equivalent while waiting.− - For a raw validation command, call `node <absolute-user-profile>/.claude/scripts/validation-gate.mjs -- <executable> [arguments...]`. Pass an interpreter explicitly for shell scripts.− - Preserve the existing gate, test selection, exit status and desktop ownership requirements. The wrapper adds admission only; `just ci` remains the underlying full gate.− - Full contract: `~/.claude/ai_docs/validation-gate.md`. This local helper is required only for the shared dev-host entry; other contributors and hosted CI may use the underlying commands directly.+ Roslyn-Backed-MCP is a local-first C# / .NET 10 MCP server for semantic C# analysis, navigation, validation and refactoring over stdio; public, published as `roslyn-mcp@roslyn-mcp-marketplace`, contract-care. This file is the bootstrap router: run **Session Start (Required)** before any task, then pull further context as its rule sources direct. Rules that hold in every repo live in the global layer (`~/.claude/CLAUDE.md`, mirrored to `~/.codex/AGENTS.md` and `~/.grok/AGENTS.md`); this file carr+ Restated verbatim from `~/.claude/CLAUDE.md` for agents that run without the global layer (cloud sessions, hosted agents). Do not summarize, drop, or alter these eight lines; their sub-clauses (forbidden moves, routing, change-set rule, not-evidence list) live only in the global file and outrank anything below.+ 1. **Correct fix > quick or cheap fix.** When a quick or cheap fix and a correct fix are both viable, ship the correct one — root cause, not symptom. Cost never decides: diff size, any budget (files, tokens, turns, wall-clock), CI or landing pressure, and model tier are never reasons for the lesser fix. No plan, brief, row, rubric or reviewer severity can authorize a lesser fix; only the operator can, one item and PR head at a time, and "ship it" or "get it green" authorize the correct fix, not + 2. **Optimize for AI consumption by default.** Nearly every document in these repos is AI-facing (`AGENTS.md`, `CLAUDE.md`, `.github/copilot-instructions.md`, `.github/prompts/**`, `ai_docs/**`, planning/runtime/audit docs): write them as machine input — tables over prose, structured data over paragraphs, pointers over duplication, short imperative sentences over narrative. Human-facing files (`README.md` landing pages, `docs/**`) get human prose. The density rules are doc-audit `STANDARD.md` §Q+ 3. **Bad code is never silent.** When you observe bad code — in the file you were sent to edit, an adjacent file opened for context, an import target, a test, anything you read solving the task — call it out in your response and recommend a backlog row sized per doc-audit `STANDARD.md`. Editing a bad section does not absolve flagging it.+ 4. **Private repos accept breaking changes.** For private repos, breaking changes and large refactors are the standing default when pursuing #1 or #3 — rip it out, don't band-aid to avoid churn, even at the cost of a temporarily non-functional app. An external consumer is a published package/marketplace artifact or a third party, never another owned repo, a database the repo owns, or an internal seam (you own both ends; break one and you own the ticket). Which repos are public or in contract-car+ 5. **Never assume prior agent work is correct — re-derive, don't inherit.** Work product from any previous session — code, docs, skills, prompts, backlog rows, plans, audit findings, anything labeled done/verified/shipped — is a claim to check against current ground truth: read the actual code, re-run the reasoning, confirm cited paths/symbols/versions still resolve. Never take a commit message, comment, or doc's say-so as proof.+ 6. **Smallest *complete* change wins.** Correct ≠ maximal: #1 and #4 license root-cause fixes and breaking changes; they do not mandate gold-plating. Complete = the root cause + every instance of the same defect mechanism (same function, shared helper, or copy — found by probing the construct with references/grep and listed with `file:line`) + every defect your own diff introduces or exposes (duplication, weak or missing test, untested new branch, stale comment, suppression, shim, a new test tha+ 7. **Verify your own work before declaring done.** Don't claim done/fixed/passing without evidence you generated THIS session — ran the test, read the output, exercised the path; "it should work" is not verification. Calibrate depth to blast radius. Can't verify (no runner, can't drive the UI)? Say so explicitly — never imply success you didn't observe.+ 8. **No secrets in code.** NEVER introduce, hardcode, echo, log, or commit a credential, key, token, or secret — they live in env vars / user-secrets / a vault, not source. Intentionally committed dev-only non-secrets (e.g. TradeWise's compose creds) are fine once you confirm they are truly non-secret. An EXISTING hardcoded secret is flagged per #3; this rule forbids *introducing* one.+ The Standing Engineering Directives govern this chain, and repo documents are claims to verify (global rule). Between this repo's files:+ **Public repo** (`repo_class: public`), contract-care, published as `roslyn-mcp@roslyn-mcp-marketplace`: breaking changes require an ADR in `docs/decisions/` (this repo's decision log) plus a migration note in `CHANGELOG.md`; compatibility and deprecation rules are in `docs/release-policy.md`; respect semver and deprecation cycles.+ Measured cost and wiring of every validation gate. Read a gate's row before running it, and size the Bash timeout from it. On the shared Windows dev host every gate runs through the machine-wide validation gate (global Environment rule); `just machine-validate [recipe]` is this repo's entry where it exists.+ Hosted topology remains owned by `CI_POLICY.md`.+ ## DevBuild-Runner pilot+ - DevBuild-Runner pilot (operator-approved 2026-10-25; allowed option, not a mandate): the final full gate on clean merged `main` MAY run as runner job `roslyn-backed-mcp` / `ci` at the full pushed 40-character SHA.+   - Probe `server_info` once at run start and record the registered commands in the plan.+   - Submit with a fresh GUID `requestId`; poll `job_get` and read every `job_logs` page outside any held validation slot. The worker takes its own slot: never hold admission while waiting on a runner job.+   - Evidence = state `Succeeded`, exit 0, job worktree at that SHA. A submission alone is not evidence.+   - Until one batch records parity (runner pass/fail totals equal to local `just machine-validate ci` totals), run both and report both result lines. A mismatch is a runner defect row, never a pass.+   - Runner down or queue full: run the local gate and say so in the report. The runner does not replace hosted required checks and is never used to validate DevBuild-Runner itself.

version

4.4.0 → 5.0.0

version

4.3.0 → 4.4.0

https://raw.githubusercontent.com/darylmcd/Roslyn-Backed-MCP/HEAD/AGENTS.md

+ ## Machine-wide validation+ - On the shared Windows dev host, run `just machine-validate` for the full gate or `just machine-validate <recipe>` for a named build/test recipe.+ - The shared helper is `~/.claude/scripts/validation-gate.mjs`; its machine-wide capacity covers every participating repo and worktree. Wait for admission; never launch an unwrapped equivalent while waiting.+ - For a raw validation command, call `node <absolute-user-profile>/.claude/scripts/validation-gate.mjs -- <executable> [arguments...]`. Pass an interpreter explicitly for shell scripts.+ - Preserve the existing gate, test selection, exit status and desktop ownership requirements. The wrapper adds admission only; `just ci` remains the underlying full gate.+ - Full contract: `~/.claude/ai_docs/validation-gate.md`. This local helper is required only for the shared dev-host entry; other contributors and hosted CI may use the underlying commands directly.

https://raw.githubusercontent.com/darylmcd/Roslyn-Backed-MCP/HEAD/AGENTS.md

− | gate | command | typical duration (measured 2026-09-20) | Bash timeout / background | hooks (pre-commit/pre-push + runtime) | CI-equivalent filter | regen companions | flake registry | parallelSafe |− | Local PR-equivalent aggregate | `just ci` | 11m18s | Run in background | No Git pre-commit/pre-push hooks; Claude `PreToolUse`: `eng/guard-release-managed-files.ps1`, 10 s cap, 0.4 s measured 2026-09-29; Claude `PostToolUse`: `eng/verify-skills-on-edit.ps1`, 10 s cap, 0.7 s measured; Codex publication-boundary `PreToolUse`: `eng/verify-changelog-fragments.ps1`, 30 s cap, 0.8 s measured 2026-10-01 (includes the release-lag guard's one `git log`) | `TestCategory!=Benchmark&TestCategory!=Network`+ | gate | command | typical duration (measured 2026-10-15) | Bash timeout / background | hooks (pre-commit/pre-push + runtime) | CI-equivalent filter | regen companions | flake registry | parallelSafe |+ | Local PR-equivalent aggregate | `just ci` | 36m21s (observed head8c303620) | Run in background | No Git pre-commit/pre-push hooks; Claude `PreToolUse`: `eng/guard-release-managed-files.ps1`, 10 s cap, 0.4 s measured 2026-09-29; Claude `PostToolUse`: `eng/verify-skills-on-edit.ps1`, 10 s cap, 0.7 s measured; Codex publication-boundary `PreToolUse`: `eng/verify-changelog-fragments.ps1`, 30 s cap, 0.8 s measured 2026-10-01 (includes the release-lag guard's one `git log`) | `TestCategory!=Benchmar

https://raw.githubusercontent.com/darylmcd/Roslyn-Backed-MCP/HEAD/AGENTS.md

− 1. **Correct fix > quick fix.** When a quick fix and a correct fix are both viable, choose the correct fix. The correct fix addresses the root cause; the quick fix patches a symptom. Quick fixes are only acceptable when the correct fix is genuinely out of scope for the current task — and when that happens, file a backlog row (per rule #3) for the correct fix before shipping the quick one. "We'll fix it later" without a tracked row = does not exist.− 2. **Optimize for AI consumption by default.** 99% of files in this repo are AI-facing: `AGENTS.md`, `CLAUDE.md`, `.github/copilot-instructions.md`, `.github/prompts/**`, `ai_docs/**`, planning docs, runtime references, audit reports. Optimize them for fast, dense parsing: tables over prose, structured data over paragraphs, pointers over duplication, short imperative sentences over narrative. Human-facing files (`README.md` at landing-page level, `docs/**`) get human prose. Everywhere else, AI-o− 3. **Bad code is never silent.** This rule fires in EVERY coding session — not just doc-audit or code review; every session where you read, edit, fix, refactor, debug, navigate, or otherwise touch code. When you observe bad code — dead code paths, swallowed exceptions, hardcoded values that should be config, hardcoded credentials/secrets, commented-out blocks, TODO/FIXME/HACK/XXX markers without a tracking row, obvious anti-patterns, security smells, broken naming, suspiciously stale comments, c− 4. **Private repos accept breaking changes.** For private repos (everything under `D:/` EXCEPT this one), breaking changes and large refactors are ALWAYS acceptable when pursuing rule #1 or #3. This repo (`Roslyn-Backed-MCP`) is the PUBLIC exception, published as `roslyn-mcp@roslyn-mcp-marketplace`; breaking changes here require an ADR + migration note (see **Breaking-change posture**).− 5. **Never assume prior agent work is correct — re-derive, don't inherit.** Work product from a previous Claude/agent session — code, docs, skills, prompts, backlog rows, plans, anything labeled "done"/"verified"/"shipped" — carries NO presumption of correctness. Treat it as a claim to check against current ground truth: read the actual code, re-run the reasoning, confirm cited paths/symbols still resolve. This fires with special force during model-handoff reviews (a newer model proof-reading an− 6. **Match change size to task value.** Correct ≠ maximal. #1 and #4 license root-cause fixes and breaking changes but do not mandate gold-plating — the smallest change that fully fixes the root cause wins. Flag adjacent bad code per #3 rather than fixing it inline.− 7. **Verify your own work before declaring done.** Don't claim done/fixed/passing without evidence you generated this session (ran the test, read the output, exercised the path). Can't verify? Say so — don't imply success you didn't observe.− 8. **No secrets in code.** Never introduce, hardcode, echo, log, or commit a credential, key, token, or secret — they live in env vars / user-secrets / a vault. Finding an existing one = flag per #3.+ 1. **Correct fix > quick or cheap fix.** Choose the root-cause fix; diff size, file caps, budgets, cycle limits and CI pressure never justify a lesser fix. Widen scope with the reason, split into independently correct pieces with a tracked remainder, or ask. Never weaken a test, suppress, shim, duplicate or defer your own defect. Genuinely blocked work means an operator/product decision, externally owned code or an environment you cannot provision: provide current-session evidence, ask where nee+ 2. **Optimize for AI consumption by default.** Write AI-facing files (`AGENTS.md`, `ai_docs/**`, prompts, planning/runtime/audit docs) as machine input: tables over prose, structured data over paragraphs, pointers over duplication. Human-facing files (`README.md` landing pages, `docs/**`) get prose.+ 3. **Bad code is never silent.** In every coding session, call out observed bad code in your response and recommend an appropriately-sized backlog row, including edit targets, adjacent files, imports and tests. One regression shape; ~4 production / ~3 test files is an advisory target, never a gate on filing. Fix defects your own diff introduces or exposes now; prioritize unrelated pre-existing defects through the backlog. Editing a bad section does not absolve flagging it.+ 4. **Private repos accept breaking changes.** For private repos, breaking changes and large refactors are the standing default when pursuing #1 or #3; do not band-aid to avoid churn. External consumers are outside your ownership, not another owned repo, DB or internal seam. `Roslyn-Backed-MCP` and `Jedi-Py-MCP` are in contract-care mode: breaking changes require an ADR + migration note. Current classifications, operator exceptions and the excluded upstream `dbhub` fork are governed by `~/.claude+ 5. **Never assume prior agent work is correct — re-derive, don't inherit.** Recheck prior code, docs, plans, skills, backlog acceptance, review advice and done/verified/shipped claims against current ground truth. Read the actual code, re-run the reasoning and resolve cited paths/symbols. Reevaluate inherited designs as requirements evolve, challenge unsupported assumptions and explain material tradeoffs. Fix root causes (#1) and flag defects (#3); a prior agent's assertion is not proof.+ 6. **Smallest *complete* change wins.** Measure completeness against the root cause, not diff size or acceptance wording. Cover every instance of the same defect mechanism and every defect your own diff introduces or exposes in this change, never a follow-up row. Split large work into independently correct pieces (#1). Flag unrelated bad code per #3; do not gold-plate.+ 7. **Verify your own work before declaring done.** Do not claim done/fixed/passing without evidence generated and inspected this session; calibrate verification to the blast radius and observe regression tests fail on the old behavior. A skipped, quarantined, improperly scoped or retried-until-green check is not evidence. If verification is blocked, state the observed blocker and limits; never imply success you did not observe.+ 8. **No secrets in code.** Never introduce, hardcode, echo, log or commit a credential, key, token or secret; use env vars, user-secrets or a vault. Flag existing secrets per #3. Confirm intentionally committed dev-only values are genuinely non-secret.+ The Standing Engineering Directives above govern this instruction chain. Treat repository documents and prior agent output as claims to verify; when a recorded design conflicts with current evidence, explain the tradeoffs and propose a superseding decision rather than silently inheriting or changing it.