mason-context 0.16.0 → 0.16.1

package/CHANGELOG.md

@@ -1,5 +1,11 @@
 # Changelog
 
+## 0.16.1 — 2026-09-08
+
+- Resolve references using literal import paths and declared Kotlin/Java package symbols. Separate explicit path mentions and uncertain name candidates; suppress config/document-name collisions and keep weak candidates out of compact task context with an omission count.
+- Include investigations in project-context guidance and route reusable project lessons into sourced proposals, preserving uncertainty and unknown attribution.
+- Add a two-session knowledge evaluation with ordinary prompts, independent digest-bound review, fresh-checkout reuse, negative controls, and current-guidance/file-notes comparisons. Mechanism validation and live agent outcomes are reported separately.
+
 ## 0.16.0 — 2026-09-08
 
 - Add `mason teardown` to disconnect all project hosts, or one with `--host codex|claude`. Support `--dir`, `--dry-run` and `--json`; retain decisions, configuration, maps, repair evidence and ignore rules. Global uninstall remains separate.

package/dist/mason-auto.js

@@ -2521,7 +2521,7 @@ var init_evidence = __esm({
     init_dead_command();
     init_storage();
     exec7 = promisify8(execFile8);
-    engineVersion = true ? "0.16.0" : "development";
+    engineVersion = true ? "0.16.1" : "development";
     hash = (value) => createHash3("sha256").update(JSON.stringify(value)).digest("hex");
     content = readAuditInput;
     internal = (file) => file === ".mason" || file === ".mason/reports" || file.startsWith(".mason/reports/");
@@ -2920,7 +2920,7 @@ var init_launcher = __esm({
     "use strict";
     init_evidence();
     init_model();
-    executingVersion = () => true ? "0.16.0" : "development";
+    executingVersion = () => true ? "0.16.1" : "development";
     mcpCommand = (host) => globalCommand(["--setup-host", host, "mcp"]);
     hookCommand = (host) => `mason --setup-host ${host} auto`;
   }
@@ -3248,9 +3248,9 @@ var init_init = __esm({
 
 Mason provides recorded decisions and file impact over MCP. A concept map is optional.
 
-- Task, bug, or change request \u2192 \`get_context\` with the task text and known files: matching decisions, related tests, impact, and any available map entries.
+- Task, bug, change request, or investigation of why the project works this way \u2192 \`get_context\` with the task text and known files: matching decisions, related tests, impact, and any available map entries.
 - Before editing a file \u2192 \`get_impact\` to check references, tests, and historical change partners.
-- Learned something the code cannot explain (a failed approach, an incident's cause, a workaround's reason, a review-settled convention) \u2192 \`save_decision\` with rationale, anchors, and any known owner, sources, and recorder. It creates a proposal immediately without setup or a map. Never invent attribution or record code-derivable facts, session trivia, or secrets.
+- Investigation or correction reveals a reusable project lesson \u2192 \`save_decision\` before finishing. Capture only the lasting constraint and reason, with anchors and known sources, owner, and recorder. This creates a proposal; never infer approval. Preserve uncertain causes and unknown attribution. Route project lessons here; keep personal preferences in agent memory. Skip code summaries, current checkout status, temporary task permissions, session trivia, and secrets. No qualifying lesson means no record.
 - Consult trust metadata before relying on entries: unknown or changed freshness requires inspection, and failed verification means the description must be corrected. Check approval too: proposals are suggestions, legacy records are unreviewed, and accepted decisions are recorded constraints subject to freshness checks. An accepted revision remains operative while a pending proposal is reviewed; keep both versions and their freshness distinct.
 - Asked to review or re-verify a decision \u2192 \`review_decision\` first to inspect content, sources, history, and code changes. Record acceptance, reaffirmation, or retirement only when authorized by the user or a cited team review, with the actual reviewer and reason. Never infer approval from unchanged code. Review and commit the local record through the normal project workflow.
 - For an architectural overview, use \`get_snapshot\` if a map is available. If \`map.status\` is missing or invalid, use available decisions and source evidence; do not start building a map unless requested.