AX-1694 — Align VS Code copilot-instructions with Claude/Cursor agent-guard docs by MatanEden1 · Pull Request #21 · jfrog/vscode-plugin

MatanEden1 · 2026-06-15T18:40:36Z

Jira Ticket Path: https://jfrog-int.atlassian.net/browse/AX-1694

The VS Code MCP-management instructions had drifted from the updated Claude and Cursor agent-guard instructions, and the plugin still shipped them via the legacy platform-specific shell scripts. This PR aligns the template with the same structure and rules as Claude/Cursor and replaces the shell scripts with a single cross-platform Node injector, while preserving the VS Code-native mechanics that intentionally differ from Claude/Cursor.

Changes

Replace the shell scripts with a Node injector. Deleted scripts/ensure-instructions.sh and scripts/ensure-instructions.ps1; added scripts/inject-instructions.mjs. The injector resolves JFrog credentials (env vars JFROG_URL/JF_URL + token, falling back to the JFrog CLI's default server via jf config export), checks whether Agent Guard is enabled for the account, and — when enabled — emits the template via hookSpecificOutput.additionalContext for the SessionStart hook. Honors _JF_AGENT_GUARD_FORCE_DISABLE / JF_AGENT_GUARD_FORCE_ENABLE. Fails closed (emits {} and exits 0) on any error.
Delivery is additionalContext-only. The injector no longer writes a workspace instructions file; the instructions are delivered solely through the SessionStart hook payload, which the IDE consumes directly.
Update hooks.json to the SessionStart shape and point it at node inject-instructions.mjs (timeout 7s).
Restructure the template (templates/jfrog-mcp-management.md) to match the Claude/Cursor layout: Pre-flight section (live-execution rule, mandatory <PROJECT>, auto-resolvable <SERVER_ID>, network note), catalog-first "Adding an MCP" flow, numbered Steps 1–5 with explicit start+verify (4a) and OAuth login (5), the Listing routing table, and expanded Key Rules / Troubleshooting. Wording that had drifted is now aligned verbatim with the references.
Preserve the VS Code secrets model: top-level inputs + ${input:...} substitution + OS-keychain prompt-on-first-start (not shell env-var export), CodeLens / MCP: List Servers mechanics, the real VS Code MCP config locations (workspace .vscode/mcp.json and the user-profile .../Code/User/mcp.json), and orphaned-inputs cleanup on removal.
Bump version to 1.0.4 (marketplace.json and plugin/.claude-plugin/plugin.json) so the new instructions ship.
Add a smoke test + CI. scripts/validate-template.mjs (run by a Validate Template workflow on every PR) asserts the injector parses, force-enable emits valid JSON with a non-empty hookSpecificOutput.additionalContext, force-disable emits {}, the template the injector reads resolves, and hooks.json wires SessionStart to the injector — guarding the template-filename/read-path drift that recurred during review. Mirrors the validators in claude-plugin/cursor-plugin.

Runtime note

For CLI-credential users (no JFROG_URL/JFROG_ACCESS_TOKEN set), the injector spawns jf config export (bounded by a 3s timeout) plus one settings network call on every SessionStart; env-var users skip the jf spawn. All paths fail closed (emit {}, exit 0).

Intentional divergences from Claude/Cursor

These differ on purpose because the IDE requires it or by product decision — they are not drift:

Default install scope is user-level (.../Code/User/mcp.json), with the workspace .vscode/mcp.json as the "for this project / commit / share" opt-in. Claude/Cursor default to the project file; VS Code defaults to user-level so installed MCPs follow the user across all workspaces.
No agent-runnable mcp list. Claude/Cursor have CLI list commands (claude mcp list / cursor agent mcp list); VS Code's MCP: List Servers is UI-only, so "Currently installed" reads the config files directly and defers live status to a user-ask.
Start & verify are user actions. Starting the server, entering inputs at the keychain prompt, and reading server status are VS Code UI actions the agent cannot perform, so 4a asks the user and acts on what they report.
The inputs/${input:...}/keychain model replaces Claude/Cursor's shell ${VAR}/${env:VAR} exports entirely.

align to:
claude-plugin commit_id: 97e25cc7db106c7fb6e2343968cbfe8fe5e5963a
cursor-plugin commit_id: 72bd39a7c2cec02fea39dbe97e51f039cc4bada7

Tests

Automated: scripts/validate-template.mjs runs on every PR via the Validate Template workflow (verified it goes red on a simulated template-filename drift and green once fixed).

Manual: exercised the injector across force-disable / force-enable / env-var creds / jf config export decode / malformed-token / missing-jf paths — all emit valid JSON and fail closed. Template reviewed section-by-section against the Claude/Cursor references.

🤖 Generated with Claude Code

…-guard docs Bring the VS Code MCP management instructions up to the structure of the updated Claude and Cursor agent-guard docs, while keeping VS Code-native mechanics: - Add Pre-flight section (live-execution/no-local-resolution rule, mandatory <PROJECT>, auto-resolvable <SERVER_ID>, network/sandbox note) - Catalog-first "Adding an MCP" flow; numbered Steps 1-5 with explicit start+verify (4a) and OAuth login (5) - Keep VS Code secrets model: top-level `inputs` + `${input:id}` + keychain prompt-on-first-start (NOT env-var export) - Routing table for Listing MCPs; available vs installed - New Managed Settings section: block non-agent-guard installs and removal of protected MCPs - Explicit project re-sync on update/switch; orphaned-input cleanup on removal; expanded Key Rules and Troubleshooting Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…ror project-switch priority into Pre-flight - "live execution" exemption now says "local config files" (covers both workspace .vscode/mcp.json and the user-level MCP config) instead of naming only the workspace file - mirror the explicit project-switch priority into the Pre-flight <PROJECT> bullet so it also governs listing/inspecting, not just writes Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Regression check vs the Claude/Cursor sources found the only dropped canonical detail: both gate --login on a remote section *with type:"http"*. VS Code had generalized it to "a remote section", which could fire --login for non-http remotes. Restore the qualifier. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…ling

YoniMelki

Review: requesting changes

Reviewed against the canonical Claude (claude-plugin/templates/jfrog-mcp-management.md) and Cursor (cursor-plugin/plugins/jfrog/templates/jfrog-mcp-management.md) templates, since the goal of this PR is cross-plugin alignment.

The alignment work itself is solid — structure, command blocks, Pre-flight resolution chains, and Key Rules all track the other two templates, and the VS Code-native mechanics (servers / inputs / ${input:...} keychain model, CodeLens, MCP: List Servers) are preserved correctly. Nice work.

Requesting changes for the items below — details in the inline comments.

Must fix (this file):

Trailing whitespace, plus a . For HTTP → .For HTTP missing-space regression in the Step 4 inputs rules.

Needs a decision / release step:

Version bump: this changes the template that ships to users, but marketplace.json (currently 1.0.2) isn't bumped. Per our release convention vscode-plugin bumps marketplace.json only — please bump it so the new instructions actually ship.
A few additions here are ahead of / diverge from Claude & Cursor (the ## Managed Settings section, the <PROJECT> re-sync Exception: clause, and the richer Step 2 error handling). Each is good on its own, but for true alignment we should either back-port them to the other two plugins or mark them as intentional VS-Code-only. Please also correct the PR description line about VS Code 'missing ... managed-settings enforcement' — it's actually ahead there.

Confirmed separately: the registry env var JFROG_AGENT_GUARD_REPO is correct (matches Claude). Cursor's JFROG_MCP_GATEWAY_REPO is the stale outlier and can be fixed in a separate change.

Address YoniMelki's review on PR #21: - Remove the VS-Code-only <PROJECT> switch/re-sync Exception clause from the Pre-flight bullet (not present in Claude/Cursor). - Reword the broken server-resolution prose into a single clear sentence. - Fix two regressions in the Step 4 inputs rules: trailing whitespace and a missing space ("${input:<id>}".For -> . For). - Remove the net-new Managed Settings section (and its cross-reference in Removing an MCP); not present in Claude/Cursor. Step 2's richer not-allowed/near-miss handling is kept intentionally (to be back-ported to Claude/Cursor for alignment). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…js injector Consolidate ensure-instructions.sh and ensure-instructions.ps1 into a single inject-instructions.mjs that works on all platforms. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

…ions Add resolveCredentials() with strict priority chain: --server flag, env vars, CLI config default profile, and single-profile fallback. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

YoniMelki

Re-review (head `3585b81`): still requesting changes

Thanks for the updates. Confirmed resolved from the previous round: the . For HTTP space regression and the trailing whitespace in the template are fixed, and the project-switch priority is now mirrored into Pre-flight.

The scope grew to replace the shell scripts with inject-instructions.mjs, and that new file has blocking issues — details inline. Summary:

Blocking

inject-instructions.mjs reads the wrong template filename (jfrog-mcp-management.md vs this repo's copilot-instructions.md) and fails silently → nothing is ever injected.
The injector no longer writes .github/copilot-instructions.md, which is the file VS Code / GitHub Copilot actually reads — only the Claude-Code additionalContext is emitted now. Need confirmation this delivery-model change is intentional.

Should fix

Asymmetric force-flag env-var names (disable has a leading _, enable doesn't).
--server resolution is dead code, since the hook invokes the script with no args.
New per-session network call on SessionStart (minor; please note it in the PR).

Process

marketplace.json is 1.0.3 on both main and this branch, so the PR doesn't bump it — confirm these changes ship under a fresh version (if 1.0.3 is already published, this needs 1.0.4).
The PR description is now stale: it doesn't mention the script→Node-injector replacement (the largest code change here), and still claims VS Code was 'missing … managed-settings enforcement while the other two plugins moved ahead' (VS Code is actually ahead). Please update both.

- Fix template filename to copilot-instructions.md (was injecting nothing) - Restore .github/copilot-instructions.md write for Copilot delivery - Align force-enable env vars with underscore-prefixed convention - Remove unreachable --server resolution branch and fix precedence docs - Lower gateway settings-check timeout from 5s to 3s Co-authored-by: Cursor <cursoragent@cursor.com>

--login launches the system browser and runs a local OAuth callback server, so a sandboxed run needs unrestricted permissions. Co-authored-by: Cursor <cursoragent@cursor.com>

YoniMelki

Re-review (head `5cc4d1d`): one alignment blocker left

Thanks for the quick turnaround — most of the last round is resolved.

Resolved

Template filename fixed to copilot-instructions.md (was injecting nothing); the read catch now logs instead of failing silently.
.github/copilot-instructions.md write restored (write-when-absent, non-fatal) — Copilot delivery is back, with additionalContext still covering Claude Code.
--server dead-code branch removed and the precedence docstring corrected.
Settings-check timeout lowered to 3s with an explanatory comment.

Blocking (alignment)

The force-flag rename diverges from Claude/Cursor — see inline. Important correction to my previous comment: the disable/enable asymmetry (_JF_AGENT_GUARD_FORCE_DISABLE with underscore, JF_AGENT_GUARD_FORCE_ENABLE without) is the intentional, identical convention in both reference plugins. My earlier 'align them' note was wrong for this PR's goal. Adding the underscore to the enable flag means JF_AGENT_GUARD_FORCE_ENABLE (the name that works in Claude/Cursor) now silently no-ops in VS Code. Please match the references exactly.

Minor — VS-Code-ahead deltas (not blocking; reconcile or confirm intentional)

resolveCredentials() adds a JF-CLI-config fallback the references don't have (they're env-only) — nice enhancement, but it changes when the settings check fires vs Claude/Cursor. Worth back-porting for parity or noting as deliberately ahead.
3s vs the references' 5s timeout, and isGatewayEnabledViaSettings / 'gateway' log wording vs the references' 'agent guard' naming (post-rename). Cosmetic.
The clearer --login sandbox-permissions note is a good candidate to back-port to Claude/Cursor.
.github/copilot-instructions.md is create-only — see inline lifecycle note.

Process

marketplace.json is 1.0.3 on both main and this branch. Per this repo's convention that's the file bumped per ticket, so AX-1694 should ship as 1.0.4 if 1.0.3 is already released.
The PR description is stale: it still lists 'Add a Managed Settings section' under Changes, but that section was removed in 0606423; and it doesn't mention the script→Node-injector replacement (the largest change here). Please update both.

- inject-instructions.mjs: fall back to ~/.jfrog/jfrog-cli.conf.v6 (default profile, else the sole profile) when JFROG_URL/token env vars are absent. - copilot-instructions template: default install location to ~/.vscode/mcp.json, clarify remote MCPs use the catalog name (no spec.packageName), and scope the inputs id uniqueness to the config file it lives in. Co-authored-by: Cursor <cursoragent@cursor.com>

Rework instructions to clarify precedence and locations for MCP configuration: prefer workspace .vscode/mcp.json as the default target, and use the VS Code user-level MCP config only for global/personal installs. Update Server ID guidance to read the server value from the workspace file or the VS Code user config (via "MCP: Open User Configuration"). Add explicit OS paths for the user-level config and emphasize writing only to the chosen scope (workspace vs user). Also update Step 4 to reference .vscode/mcp.json consistently and remind that secrets must use ${input:...} substitution.

…ttings

Consolidate all instruction references to use only ~/.vscode/mcp.json and remove the workspace .vscode/mcp.json branches for a single, consistent config location. Co-authored-by: Cursor <cursoragent@cursor.com>

VS Code never reads ~/.vscode/mcp.json. Point the instructions at the two locations it actually loads: the workspace .vscode/mcp.json and the user-profile mcp.json (MCP: Open User Configuration), with the correct per-OS paths, and explicitly warn off ~/.vscode/mcp.json. Co-authored-by: Cursor <cursoragent@cursor.com>

Update instructions to refer to the user-level ~/.vscode/mcp.json as the default

YoniMelki

Re-review (head `af832bb`): injector regressed — please don't ship this version

The template MCP-config work this round is good (see below), but inject-instructions.mjs was replaced with what looks like a verbatim copy of the Cursor injector, which re-broke two blockers from earlier rounds and added a third. As it stands on af832bb, the injector delivers nothing to either Copilot or Claude Code. Details inline; summary:

Blocking (all in inject-instructions.mjs)

Wrong template filename, again — reads templates/jfrog-mcp-management.md, which does not exist in this repo (the only template is copilot-instructions.md). The read throws, the catch writes {} and exits → nothing injected. This is the round-1 critical bug, reintroduced.
.github/copilot-instructions.md write removed, again — the file-materialization block that was restored last round is gone, so Copilot has no instructions file to read. Round-1 major regression, reintroduced.
Wrong output key for this hook — emits top-level additional_context (Cursor's format), but hooks.json is a Claude Code SessionStart hook, which reads hookSpecificOutput.additionalContext. Even with the filename fixed, the payload would be ignored.

Resolved / good

Force-flag fixed: forceEnabled is now JF_AGENT_GUARD_FORCE_ENABLE (no underscore), matching Claude/Cursor. Thanks for the correction.
resolveCredentials() refactor is cleaner and keeps the env→JF-CLI-config fallback; naming aligned; 5s timeout matches the references.
Template config-location work is accurate: it correctly warns against ~/.vscode/mcp.json and points to the real user-profile paths (…/Code/User/mcp.json) and the workspace .vscode/mcp.json. Nicely done.

The fix: keep this round's force-flag fix and refactor, but restore the three VS-Code-specific bits the Cursor copy dropped — read copilot-instructions.md, re-add the .github/copilot-instructions.md write, and emit hookSpecificOutput.additionalContext. Effectively the prior head's injector plus this round's force-flag change.

Process

marketplace.json is still 1.0.3 on both main and this branch — no bump; should be 1.0.4 per the per-ticket convention if 1.0.3 is released.
PR description is still stale: it lists 'Add a Managed Settings section' (removed in 0606423) and omits the script→Node-injector rewrite.

Fixed template path to copilot-instructions.md. Restored automatic workspace write to .github/copilot-instructions.md if missing. Aligned stdout payload with Claude Code's SessionStart hook format. Added missing fs imports and explicit debug error logging.

YoniMelki

Re-review (head `dace1a6`): injector fixed ✅ — but a new template regression

Thanks Matan — all three injector blockers are resolved and verified:

reads copilot-instructions.md (not the nonexistent jfrog-mcp-management.md);
restores the .github/copilot-instructions.md write (write-when-absent, non-fatal);
emits hookSpecificOutput.additionalContext, the correct shape for this Claude Code SessionStart hook.

The injector is now coherent and correct. However, a separate commit this round (ba067fa, "Clarify user-level MCP config path") regressed the template in the wrong direction, and it's blocking.

Blocking — wrong user-level MCP config path (7 places)
ba067fa replaced the user-level config path with ~/.vscode/mcp.json everywhere (lines 87, 121, 123, 197, 248, 324, 358) and deleted the correct paths. Per the VS Code docs, the user-level mcp.json lives in the user-profile folder — ~/Library/Application Support/Code/User/mcp.json (macOS), ~/.config/Code/User/mcp.json (Linux), %APPDATA%\Code\User\mcp.json (Windows). VS Code does not read an mcp.json in ~/.vscode/ (that folder holds extensions and argv.json). The previous head (af832bb) had this correct and even warned against this exact path — please restore those paths. Details inline. (Refs: https://code.visualstudio.com/docs/copilot/chat/mcp-servers and https://code.visualstudio.com/docs/agent-customization/mcp-servers)

Process (unchanged)

marketplace.json is still 1.0.3 on both main and this branch — no bump; should be 1.0.4 per the per-ticket convention.
PR description is still stale: it lists 'Add a Managed Settings section' (removed in 0606423) and omits the script→Node-injector rewrite.

Require reachable JFROG_URL credentials for agent guard commands

YoniMelki

Re-review at head 5036f0d.

Landed this round (good ✅) — commit 5036f0d adds the credentials block (template lines ~49–72): JFROG_URL must be the bare platform base URL, present in the command's own environment, with a clear explanation of the '<' looking for beginning of value HTML-not-JSON failure. Accurate, well-scoped, closes a real footgun.

Injector — unchanged since dace1a6 and still correct: reads copilot-instructions.md, writes .github/copilot-instructions.md when absent, emits hookSpecificOutput.additionalContext. All earlier injector blockers remain resolved. ✅

🔴 Blocker (still open) — wrong user-level MCP config path. This is the one remaining functional blocker, and it's now an open disagreement rather than an oversight: the template uses ~/.vscode/mcp.json for the user-level config in 7 places, and Matan's replies note we'd "talked about" that path. I've replied in detail on both inline threads (write path, line 144; read path, line 379) with the evidence:

VS Code's MCP: Open User Configuration opens mcp.json in the user profile folder (~/Library/Application Support/Code/User/ on macOS), per the docs — not the home ~/.vscode/ folder (which is for extensions + argv.json).
Verified on a real machine with two active user-level gallery MCPs: VS Code wrote them to ~/Library/Application Support/Code/User/mcp.json (+ a Code/User/mcp/ runtime dir); ~/.vscode/mcp.json was never created.
Net effect as written: install silently no-ops (line 144) and "Currently installed" always reads empty (line 379).

Correct user-level paths — macOS ~/Library/Application Support/Code/User/mcp.json, Windows %APPDATA%\Code\User\mcp.json, Linux ~/.config/Code/User/mcp.json. Workspace .vscode/mcp.json is correct and stays.

One-step confirmation: run MCP: Open User Configuration and read the path in the editor tab. If it's …/Code/User/mcp.json, this needs the fix; if it genuinely opens ~/.vscode/mcp.json, I'm wrong — say so on the thread and I'll drop it.

Process notes (carried forward, unchanged):

marketplace.json not bumped — plugins[0].version is still 1.0.3 on both head and main; should be 1.0.4 per convention.
PR description is stale: it lists "Add a Managed Settings section," but there's no such section in the template (the earlier ## Managed Settings heading is gone — its removal isn't reflected in the description); and it still omits the injector rewrite (ensure-instructions.sh/.ps1 → inject-instructions.mjs), the largest structural change in the PR.

- Replace wrong user-level path (~/.vscode/mcp.json) with correct platform-specific VS Code user-profile paths in 7 template locations: macOS ~/Library/Application Support/Code/User/mcp.json, Linux ~/.config/Code/User/mcp.json, Windows %APPDATA%\Code\User\mcp.json. Step 1 "Target config file" is the single source of truth; other sections back-reference it. - Add "timeout": 7 to hooks.json to match claude-plugin reference and provide a hard kill-switch above the injector's 5s fetch timeout. - Document create-only .github/copilot-instructions.md lifecycle in both the script comment and a new Troubleshooting entry so users know to start a new Claude Code session after deleting for a refresh. - Bump marketplace.json version 1.0.3 → 1.0.4 so the updated template and hook actually ship to users. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

…jection, fix token leak Apply davida-jfrog's review feedback on the VS Code agent-guard plugin: - Rename templates/copilot-instructions.md -> templates/jfrog-mcp-management.md for consistency with the Claude and Cursor plugins, and update the read path in inject-instructions.mjs to match. - Deprecate the .github/copilot-instructions.md materialization. Copilot consumes hookSpecificOutput.additionalContext directly, so writing the file as well duplicated the instructions. The SessionStart additionalContext payload is now the sole injection mechanism; remove the now-unused writeFileSync/mkdirSync/existsSync imports and the obsolete "Copilot instructions look outdated" troubleshooting bullet. - Stop instructing the agent to prepend JFROG_ACCESS_TOKEN inline on the command line (leaks the secret into shell history / process listings). Credentials now resolve from already-exported env vars or via --server reading jfrog-cli.conf.v6; only the non-secret JFROG_URL may be prepended. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…ctor comment - Drop the "Credentials must be present in the command's OWN environment / bare-JFROG_URL" Pre-flight block from the template (davida-jfrog: not VS Code-specific; the underlying /ui->HTML footgun is tracked for the shared Claude/Cursor instructions instead). Credential resolution is still covered by the <SERVER_ID> Pre-flight bullets. - Reword the injection comment to drop the now-irrelevant Copilot / copilot-instructions.md references. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…/Cursor - Bump plugin.json to 1.0.4 (in tandem with marketplace.json). - Drop the sandbox/full-network wording (VS Code has no command sandbox) in the Pre-flight, Step 5 --login note, and the 403 troubleshooting entry; reframe as network/proxy/registry guidance. - Clarify agent-vs-user actions in the "0 tools" troubleshooting entry (the agent cannot operate the VS Code UI) and add a framing note. - Revert the lines 63-68 wording to match the Claude/Cursor templates so the shared prose stays aligned. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

… CLI config Per David's review: stop hand-parsing ~/.jfrog/jfrog-cli.conf.v6 (its format varies across CLI versions). Env vars still take precedence; the CLI fallback now runs 'jf config export' (default server) and decodes the base64 Config Token for {url, accessToken}, letting the CLI own all config-format parsing. Falls back to injecting nothing when jf is absent or no server is configured. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

davida-jfrog · 2026-06-22T09:40:59Z

+Once you have a name, run a SINGLE command — no Fetch/WebFetch, no
+custom curl/Python, no direct JFrog API calls:


Why is this differnt from cursor?

already resolved

davida-jfrog · 2026-06-22T09:51:39Z

+Writing the entry to `mcp.json` is not enough — the server has
+to start and actually expose tools.
+
+1. VS Code detects the edited `mcp.json` and offers a **Start** action


The agent cant do it so this should say what the agent should ask the user to do more explicitly.

already resolved

davida-jfrog · 2026-06-22T09:58:23Z

+1. Open `MCP: List Servers` for connection status (one row per
+   server: Running / Stopped / Failed).


It looks like the agent cant execute this.

I dont think there is a need for the user to list the mcp servers manually in the normal case. the agent can do it by themselfes.

For troubleshooting - fine.

already resolved

Template: - Match Claude/Cursor verbatim where wording had drifted: Step 1 "Once chosen" server rule, Step 2 lead-in, Step 2 error handling, "Available to install" item 3, the Pre-flight live-execution bullet. - Rename the server-key placeholder PACKAGE_ID -> spec.packageName and drop the now-redundant identifier note. - Step 4 inputs example now shows one secret + one non-secret input. - 4a reframed: starting/verifying are user actions; agent asks and acts on what the user reports. - "Currently installed" reads config files directly (no agent-runnable MCP: List Servers); UI status deferred to troubleshooting. Injector: - Trim history-referencing comments per review. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

YoniMelki · 2026-06-22T12:55:22Z

+"add an MCP", "what can I install" — your FIRST action is to show
+them the catalog so they can pick:
+
+1. Resolve server (Server ID`<SERVER_ID>` or URL `JFROG_URL`)


Two carried-over typos from the references: missing space in Server ID before <SERVER_ID> here, and line 66 says "not env vars" (should be "no env vars"). They match claude/cursor so byte-equivalence holds, but they're real typos in all three - worth one cleanup pass across the plugins.

Adds scripts/validate-template.mjs and a Validate Template workflow, mirroring claude-plugin / cursor-plugin which already ship validators. The test guards the failure mode that recurred 3× on this PR: a template-filename / read-path mismatch makes the injector silently emit nothing (it catches the read error and exits 0). Checks: - injector parses (node --check) - the templates/<file> the injector reads actually exists and is non-empty - force-enable emits valid JSON with non-empty hookSpecificOutput.additionalContext - force-disable emits {} - hooks.json wires SessionStart to inject-instructions.mjs Verified it goes red on a simulated filename drift and green once fixed. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…idator comments - Bound `jf config export` with { timeout: 3000 } so it self-limits instead of relying only on the 7s hook timeout. - Write {} on the template-read failure path so all fail-closed paths are uniform. - Trim validator comments. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…sistency

…est lint - Wrap checks in a main() entry point. - Group output into Syntax / Lint / Format / Injection logic; split the force-enable check into payload-format vs injection-logic. - Add manifest lint (marketplace.json <-> plugin.json name + version match, source/hooks paths exist), adapted from YhonatanArbel's validation. - Workflow: rename to "Validate hook injection", trigger on the manifest paths the validator now reads, drop the workflow self-reference path. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Describe the user-level config as "personal, not committed" and the workspace config as "commit / share", matching the vocabulary Claude and Cursor use for the same scopes. User-level stays the default (installed MCPs follow the user across all workspaces) — only the default choice differs, intentionally. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

… running

- Server-ID resolution (Pre-flight + Step 1) now lists servers with `jf config show --format=json` instead of parsing ~/.jfrog/jfrog-cli.conf.v6; the CLI masks tokens, so the "don't print secrets" caveat is dropped. Picks the server with "isDefault": true. Keeps the two sections consistent. - 4a: drop the meta-narration; add a gate that skips steps 1-3 when the server is already enabled and running; tighten the steps. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

MatanEden1 and others added 4 commits June 15, 2026 15:30

Refine copilot instructions: clarify project switching and input hand…

a033dca

…ling

YoniMelki requested changes Jun 15, 2026

View reviewed changes

MatanEden1 and others added 3 commits June 16, 2026 10:56

AX-1694 - Replace platform-specific scripts with cross-platform Node.…

6a626ac

…js injector Consolidate ensure-instructions.sh and ensure-instructions.ps1 into a single inject-instructions.mjs that works on all platforms. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

AX-1694 - Implement full authentication precedence in inject-instruct…

3585b81

…ions Add resolveCredentials() with strict priority chain: --server flag, env vars, CLI config default profile, and single-profile fallback. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

MatanEden1 requested a review from YoniMelki June 16, 2026 11:34

YoniMelki requested changes Jun 16, 2026

View reviewed changes

Comment thread plugin/scripts/inject-instructions.mjs

Comment thread plugin/scripts/inject-instructions.mjs

Comment thread plugin/scripts/inject-instructions.mjs Outdated

Comment thread plugin/scripts/inject-instructions.mjs Outdated

Comment thread plugin/scripts/inject-instructions.mjs Outdated

MatanEden1 and others added 2 commits June 16, 2026 16:47

AX-1694 - Clarify --login requires full permissions, not just network

5cc4d1d

--login launches the system browser and runs a local OAuth callback server, so a sandboxed run needs unrestricted permissions. Co-authored-by: Cursor <cursoragent@cursor.com>

MatanEden1 requested a review from YoniMelki June 17, 2026 06:03

YoniMelki requested changes Jun 17, 2026

View reviewed changes

Comment thread plugin/scripts/inject-instructions.mjs Outdated

Comment thread plugin/scripts/inject-instructions.mjs Outdated

Comment thread plugin/scripts/inject-instructions.mjs Outdated

MatanEden1 and others added 6 commits June 17, 2026 11:21

Refine MCP configuration instructions for user-level and workspace se…

e247621

…ttings

Standardize MCP config on user-level ~/.vscode/mcp.json

2fc87e1

Consolidate all instruction references to use only ~/.vscode/mcp.json and remove the workspace .vscode/mcp.json branches for a single, consistent config location. Co-authored-by: Cursor <cursoragent@cursor.com>

Clarify user-level MCP config path

ba067fa

Update instructions to refer to the user-level ~/.vscode/mcp.json as the default

YoniMelki requested changes Jun 17, 2026

View reviewed changes

Comment thread plugin/scripts/inject-instructions.mjs

Comment thread plugin/scripts/inject-instructions.mjs

MatanEden1 requested a review from YoniMelki June 17, 2026 11:02

YoniMelki requested changes Jun 17, 2026

View reviewed changes

Comment thread plugin/templates/copilot-instructions.md Outdated

Comment thread plugin/templates/copilot-instructions.md Outdated

Update copilot-instructions.md

5036f0d

Require reachable JFROG_URL credentials for agent guard commands

MatanEden1 requested a review from YoniMelki June 17, 2026 19:06

YoniMelki requested changes Jun 17, 2026

View reviewed changes

davida-jfrog requested changes Jun 18, 2026

View reviewed changes

YoniMelki and others added 2 commits June 20, 2026 18:57

davida-jfrog requested changes Jun 21, 2026

View reviewed changes

MatanEden1 requested a review from davida-jfrog June 21, 2026 15:47

davida-jfrog reviewed Jun 22, 2026

View reviewed changes

MatanEden1 requested a review from YoniMelki June 22, 2026 12:22

YoniMelki requested changes Jun 22, 2026

View reviewed changes

YoniMelki reviewed Jun 22, 2026

View reviewed changes

Comment thread plugin/templates/jfrog-mcp-management.md Outdated

MatanEden1 and others added 4 commits June 22, 2026 18:52

Rename validation scripts and update injector checks for template con…

b96eee5

…sistency

Enhance SessionStart injector validation with plugin packaging checks

ae0ccfc

YoniMelki requested changes Jun 23, 2026

View reviewed changes

Comment thread scripts/validate-hook-injector.mjs

Comment thread .github/workflows/validate-inject-instructions.yml Outdated

Comment thread .github/workflows/validate-inject-instructions.yml

Comment thread .github/workflows/validate-inject-instructions.yml Outdated

MatanEden1 requested a review from YoniMelki June 23, 2026 10:00

YoniMelki previously approved these changes Jun 23, 2026

View reviewed changes

MatanEden1 dismissed YoniMelki’s stale review via 1dfeb0c June 23, 2026 10:11

MatanEden1 and others added 2 commits June 23, 2026 20:19

Update user instructions for server start: avoid prompting if already…

bd96899

… running

		Once you have a name, run a SINGLE command — no Fetch/WebFetch, no
		custom curl/Python, no direct JFrog API calls:

		1. Open `MCP: List Servers` for connection status (one row per
		server: Running / Stopped / Failed).

Uh oh!

Conversation

MatanEden1 commented Jun 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

YoniMelki left a comment

Choose a reason for hiding this comment

Review: requesting changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

YoniMelki left a comment

Choose a reason for hiding this comment

Re-review (head 3585b81): still requesting changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

YoniMelki left a comment

Choose a reason for hiding this comment

Re-review (head 5cc4d1d): one alignment blocker left

Uh oh!

Uh oh!

Uh oh!

Uh oh!

YoniMelki left a comment

Choose a reason for hiding this comment

Re-review (head af832bb): injector regressed — please don't ship this version

Uh oh!

Uh oh!

Uh oh!

YoniMelki left a comment

Choose a reason for hiding this comment

Re-review (head dace1a6): injector fixed ✅ — but a new template regression

Uh oh!

Uh oh!

Uh oh!

YoniMelki left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

MatanEden1 commented Jun 15, 2026 •

edited

Loading

Re-review (head `3585b81`): still requesting changes

Re-review (head `5cc4d1d`): one alignment blocker left

Re-review (head `af832bb`): injector regressed — please don't ship this version

Re-review (head `dace1a6`): injector fixed ✅ — but a new template regression