edgegate-mcp 0.2.1 → 0.2.2

package/README.md

@@ -77,7 +77,9 @@ Slash commands you can invoke directly:
 - `/edgegate-init` — full onboarding flow (zero → CI gate)
 - `/edgegate-gate` — trigger a run on an existing pipeline
 - `/edgegate-status` — check a run's status + metrics
-- `/edgegate-audit` — fetch the audit PDF for a run
+- `/edgegate-audit` — fetch the evidence bundle for a run
+- `/edgegate-compare` — diff two runs (auto-baseline) with REGRESSION/IMPROVEMENT/NEUTRAL verdict
+- `/edgegate-export` — save a run report as a markdown file (for PR comments, Slack, compliance)
 
 ## License
 

package/dist/version.d.ts

@@ -1,2 +1,2 @@
-export declare const VERSION = "0.2.1";
-export declare const USER_AGENT = "edgegate-mcp/0.2.1";
+export declare const VERSION = "0.2.2";
+export declare const USER_AGENT = "edgegate-mcp/0.2.2";

package/dist/version.js

@@ -1,3 +1,3 @@
-export const VERSION = "0.2.1";
+export const VERSION = "0.2.2";
 export const USER_AGENT = `edgegate-mcp/${VERSION}`;
 //# sourceMappingURL=version.js.map

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "edgegate-mcp",
-  "version": "0.2.1",
+  "version": "0.2.2",
   "description": "MCP server for EdgeGate — set up edge-AI regression gates from Claude Code, Cursor, or Claude Desktop.",
   "license": "MIT",
   "type": "module",

package/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "edgegate",
-  "version": "0.1.0",
-  "description": "Edge-AI regression gates from Claude Code — set up CI, run benchmarks, and fetch audit PDFs from any prompt.",
+  "version": "0.2.2",
+  "description": "Edge-AI regression gates from Claude Code — set up CI, run benchmarks, compare runs, export reports, and fetch audit bundles from any prompt.",
   "author": "Frozo / EdgeGate",
   "homepage": "https://edgegate.frozo.ai",
   "repository": "https://github.com/frozo-ai/edgegate-mcp",
@@ -21,6 +21,8 @@
     "skills/edgegate-init.md",
     "skills/edgegate-gate.md",
     "skills/edgegate-status.md",
-    "skills/edgegate-audit.md"
+    "skills/edgegate-audit.md",
+    "skills/edgegate-compare.md",
+    "skills/edgegate-export.md"
   ]
 }

package/skills/edgegate-compare.md

@@ -0,0 +1,31 @@
+---
+name: edgegate-compare
+description: Diff two EdgeGate runs in the same pipeline. Use when the user wants to know "what changed" between runs, "is this a regression", "compare these runs", or "show the delta vs main".
+---
+
+# /edgegate-compare
+
+The user wants to compare two EdgeGate runs and understand the verdict.
+
+## Steps
+
+1. **Identify the candidate run.** If the user gave you a `run_id`, use it. If they said "the latest run" or didn't specify, call `edgegate_get_report` to list recent runs and ask which one they mean.
+
+2. **Identify the baseline.**
+   - If the user provided a `baseline_run_id`, use it.
+   - Otherwise, omit the field — `edgegate_compare_runs` auto-selects the most recent PASSED run from the same pipeline (excluding the candidate). This is almost always what users want.
+
+3. **Call `edgegate_compare_runs`** with `workspace_id`, `run_id`, and optionally `baseline_run_id`.
+
+4. **Lead with the verdict.** The tool returns one of:
+   - **REGRESSION** — at least one gate flipped ✓→✗ OR a lower-is-better metric increased by ≥ 25%. Call this out at the top. List which gates flipped and what metric jumped.
+   - **IMPROVEMENT** — at least one ✗→✓ gate recovery with no regressions. Briefly highlight what got better.
+   - **NEUTRAL** — no significant changes. Reassure the user the run is safe to merge.
+   - **NO BASELINE** — this is the first run in the pipeline (nothing to compare against).
+
+5. **For PR comments:** suggest the user attach the metric deltas table + verdict line. The audit trail (signed diff SHA-256) is in the response and worth including for compliance.
+
+## Failure modes
+
+- **404 on the candidate** — wrong `run_id`. Ask the user to double-check or call `edgegate_get_report`.
+- **NO BASELINE on a pipeline that should have runs** — only one run exists in that pipeline OR the prior runs never completed. Check via `edgegate_get_report` and explain.

package/skills/edgegate-export.md

@@ -0,0 +1,34 @@
+---
+name: edgegate-export
+description: Save an EdgeGate run report as a markdown file on disk. Use when the user wants to "export", "download", "save", or "share" a run report — e.g. for a PR comment, Slack message, or compliance record.
+---
+
+# /edgegate-export
+
+The user wants the run report saved as a file they can share or attach.
+
+## Steps
+
+1. **Identify the run.** If they gave a `run_id`, use it. If not, call `edgegate_get_report` and ask which run.
+
+2. **Pick the output path.**
+   - If the user said "save to Downloads" or similar, use `~/Downloads`.
+   - If they said "save next to my code" or didn't specify, default to the current working directory (the tool defaults to `./edgegate-run-{id-short}.md`).
+   - If they said "save as <name>", use that exact filename.
+   - The tool accepts directory paths (auto-appends a filename), file paths (used as-is), `~/` (expanded to home), and relative paths.
+
+3. **Decide whether to include the diff.**
+   - If the user mentioned "for the PR" or "vs main" or "with the comparison", pass `include_diff: true` — adds the run-vs-baseline diff section to the report.
+   - Otherwise default `include_diff: false` — keeps the report focused on this run.
+
+4. **Call `edgegate_export_run_report`** with the args.
+
+5. **Confirm the file path** the tool returns and show the preview. Tell the user:
+   - The file is ready to attach to a PR comment, paste into Slack, or store with compliance records.
+   - If they want to also get a programmatic JSON version, point them at `edgegate_get_audit_report` for the bundle metadata.
+
+## Failure modes
+
+- **404 on the run** — wrong `run_id`. Check `edgegate_get_report`.
+- **In-flight run** — the report omits gate results and bundle sections (they don't exist yet) but still writes a partial file noting the status. Tell the user to retry after the run completes.
+- **Permission errors writing the file** — fall back to `/tmp/edgegate-run-{id}.md` and tell the user.