v0.36.0

What's Changed

• feat(mcp): plune mcp conformance writes JUnit XML by @AZANIR in #209
• ci(mcp): run plune mcp conformance in the MCP self-check step, from one script by @AZANIR in #207
• ci(mcp): nightly report imports conformance JUnit; unmatched tests go to review by @AZANIR in #210
• ci(mcp): call one local tool in the self-check conformance run by @AZANIR in #211
• chore(release): 0.36.0 by @AZANIR in #212

Full Changelog: v0.35.0...v0.36.0

v0.35.0

What's Changed

• ci(mcp): check our own MCP server with plune mcp lint and check by @AZANIR in #197
• docs(mcp): name a server that exits before initialize in the check exit codes by @AZANIR in #198
• test(mcp-testing): an enum that is not a list is not classified by @AZANIR in #199
• feat(eval): repeat rows with runs and pass_threshold by @AZANIR in #200
• feat(mcp): plune mcp conformance: catalog, MUST-only score, frozen profile by @AZANIR in #201
• feat(config): an mcp block in plune.yaml for agentic evals by @AZANIR in #202
• feat(mcp): run the protocol checks of plune mcp conformance by @AZANIR in #203
• fix(mcp): logging and completion checks are SHOULD; the JSON-RPC error checks are derived by @AZANIR in #204
• feat(mcp): the tool loop of an agentic eval, with a transcript per row by @AZANIR in #205
• feat(assertions): tool-call checks over the transcript of an agentic eval by @AZANIR in #206
• chore(release): 0.35.0 by @AZANIR in #208

Full Changelog: v0.34.6...v0.35.0

v0.34.6

What's Changed

• docs(mcp): how to test the MCP server, and the README count of its tools by @AZANIR in #192
• docs(mcp): a Test an MCP server section in the README by @AZANIR in #193
• test(mcp-testing): check, snapshot and diff never call the platform by @AZANIR in #194
• fix(gate): --fail-on-failures fails a run no result reached by @AZANIR in #195
• chore(release): 0.34.6 by @AZANIR in #196

Full Changelog: v0.34.5...v0.34.6

v0.34.5

What's Changed

• feat(mcp): plune mcp serve answers both protocol eras by @AZANIR in #176
• feat(mcp): lint rules for a tools/list answer, run over our own tools by @AZANIR in #177
• feat(mcp): a stable tools/list in the 2026-07-28 era, hidden tools called by name by @AZANIR in #178
• feat(mcp): a contract snapshot of our tools and a classified diff by @AZANIR in #179
• feat(mcp): a measured context budget for our tool list by @AZANIR in #180
• feat(mcp-testing): a connection core for foreign MCP servers by @AZANIR in #181
• feat(mcp): the context budget calibrated and measured on live answers by @AZANIR in #182
• feat(cli): plune mcp lint by @AZANIR in #183
• feat(cli): plune mcp check by @AZANIR in #184
• test(mcp-testing): wait for the server to be gone after plune exits on a signal by @AZANIR in #186
• feat(cli): plune mcp snapshot and plune mcp diff by @AZANIR in #185
• fix(mcp-testing): bound the pairwise lint rules by tool count by @AZANIR in #187
• fix(mcp-testing): bound the input schema walk of the lint rules by depth by @AZANIR in #188
• feat(mcp): github, markdown and junit-xml reports for plune mcp by @AZANIR in #189
• fix(mcp-testing): a stdio message over the client limit is too large, not an exited server by @AZANIR in #190
• chore(release): 0.34.5 by @AZANIR in #191

Full Changelog: v0.34.4...v0.34.5

v0.34.4

What's Changed

• fix(diff): an eval with no judged row is not measured, not a pass by @AZANIR in #174
• chore(release): 0.34.4 by @AZANIR in #175

Full Changelog: v0.34.3...v0.34.4

v0.34.3

What's Changed

• feat(mcp): plune_review_decide explains policy_refused by @AZANIR in #171
• fix(run): a JSONL dataset row that is not a row is a config error, not a crash by @AZANIR in #172
• chore(release): 0.34.3 by @AZANIR in #173

Full Changelog: v0.34.2...v0.34.3

v0.34.2

What's Changed

• feat(mcp): plune_cases_generate starts cases from pasted text by @AZANIR in #169
• chore(release): 0.34.2 by @AZANIR in #170

Full Changelog: v0.34.1...v0.34.2

v0.34.1

What's Changed

• fix(diff): an eval that was not measured does not pass the gate (plune-ai/plune#2414) by @AZANIR in #166
• fix(run): defaults.assertions run on every eval (plune-ai/plune#2415) by @AZANIR in #165
• chore(release): 0.35.0 by @AZANIR in #167
• chore(release): 0.34.1 instead of 0.35.0 by @AZANIR in #168

Full Changelog: v0.34.0...v0.34.1

v0.34.0

What's Changed

• feat(mcp): plune mcp serve and the shared tool contract by @AZANIR in #163
• chore(release): 0.34.0 by @AZANIR in #164

Full Changelog: v0.33.0...v0.34.0

v0.33.0

What's Changed

• feat(run-result): usage.priced tells a known $0 from a price nobody knows (#151) by @AZANIR in #159
• feat(import): k6 summaries and Lighthouse reports import as runs, thresholds as pass or fail (#149) by @AZANIR in #160
• test(import): an Allure result with a BOM, and README notes on k6 file names and threshold exits (#149) by @AZANIR in #161
• chore(release): 0.33.0 by @AZANIR in #162

Full Changelog: v0.32.2...v0.33.0