context-forge 0.10.0: How to Act — list assumptions up front, harden against rationalization

plugins/context-forge/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "context-forge",
   "displayName": "Context Forge",
-  "version": "0.9.0",
+  "version": "0.10.0",
   "description": "A context-engineering toolkit for AI coding agents. context-constitution reverse-engineers a repo's non-obvious tribal knowledge into an evidence-cited context-constitution.md plus a behavioral contract in CLAUDE.md; context-scrubber audits the repo's context files and reports every line that fails the litmus test — stale citations, restated facts, cross-file duplication, contradicted-by-code, should-be-just-in-time, brittle/over-specified, bloat — plus two file-level advisories, a context-budget rot check and a silent-skill (weak trigger surface) audit, with an optional gated trim. Adds the high-signal context, removes the low-signal; invents nothing, every rule and verdict is grounded.",
   "author": {
     "name": "David Castillo",

plugins/context-forge/.codex-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "context-forge",
-  "version": "0.9.0",
+  "version": "0.10.0",
   "description": "A context-engineering toolkit for AI coding agents. context-constitution reverse-engineers a repo's non-obvious tribal knowledge into an evidence-cited context-constitution.md plus a behavioral contract in CLAUDE.md; context-scrubber audits the repo's context files and reports every line that fails the litmus test — stale citations, restated facts, cross-file duplication, contradicted-by-code, should-be-just-in-time, brittle/over-specified, bloat — plus two file-level advisories, a context-budget rot check and a silent-skill (weak trigger surface) audit, with an optional gated trim. Adds the high-signal context, removes the low-signal; invents nothing, every rule and verdict is grounded.",
   "author": {
     "name": "David Castillo",

plugins/context-forge/.cursor-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "context-forge",
   "displayName": "Context Forge",
-  "version": "0.9.0",
+  "version": "0.10.0",
   "description": "A context-engineering toolkit for AI coding agents. context-constitution reverse-engineers a repo's non-obvious tribal knowledge into an evidence-cited context-constitution.md plus a behavioral contract in CLAUDE.md; context-scrubber audits the repo's context files and reports every line that fails the litmus test — stale citations, restated facts, cross-file duplication, contradicted-by-code, should-be-just-in-time, brittle/over-specified, bloat — plus two file-level advisories, a context-budget rot check and a silent-skill (weak trigger surface) audit, with an optional gated trim. Adds the high-signal context, removes the low-signal; invents nothing, every rule and verdict is grounded.",
   "author": {
     "name": "David Castillo",

plugins/context-forge/skills/context-constitution/SKILL.md

@@ -15,7 +15,7 @@ rework — into durable, complementary artifacts (two core; a third when defects
   Norms), each grounded in real evidence (multi-site citations, an authoritative site, or a commit);
   and
 - a **behavioral contract** prepended to the **root `CLAUDE.md`**: four repo-agnostic behaviors
-  (ask before assuming · minimum viable · surgical diffs · verify against a stated finish line) that
+  (ask before assuming, listing assumptions up front · minimum viable · surgical diffs · verify against a stated finish line) that
   shape *how* an agent works. It goes in the root only — it's generic and `CLAUDE.md` loads from the
   root down, so one copy is always in context; an identical copy per module is the duplication this
   tool fights (**CF-N11**); and

plugins/context-forge/skills/context-constitution/templates/behavioral-contract.md

@@ -15,15 +15,19 @@ Read this first — it governs *how* you work here; everything below is the *wha
 These four behaviors are the operating defaults; the rest of this file (and the constitution) is
 context you load per task.
 
-1. **Don't assume — ask, and surface tradeoffs.** On ambiguity, a missing detail, or a design fork,
-   stop and raise it; never paper over it with a silent guess.
+1. **Don't assume — ask, and surface tradeoffs.** Before acting, always list every assumption you
+   are making on your own — scope, interpretation, defaults, a choice between options — each with
+   its basis, or say "none". On ambiguity, a missing detail, or a design fork, stop and raise it.
+   *Red flag:* "it's obvious" / "too small to ask" — that *is* the assumption; list it.
 2. **Write the minimum that solves the stated problem.** Nothing speculative — no abstraction,
-   option, or "while I'm here" scaffolding the task didn't ask for. Would a senior engineer call it
-   overbuilt for what was requested? Then simplify.
-3. **Touch only what the task requires.** Keep diffs surgical and auditable; clean up orphans *you*
-   introduced, but don't reformat or "improve" code nobody asked you to touch.
-4. **Define success up front, then loop until verified.** State the pass condition before you start,
-   then run to it — write the check, run it, fix, re-run — and don't declare victory mid-loop.
+   option, or "while I'm here" scaffolding. *Red flag:* "might need it later", "for flexibility".
+3. **Touch only what the task requires.** Surgical, auditable diffs; clean up only orphans *you*
+   introduced. *Red flag:* "cleaner if I also…" — propose it, don't do it.
+4. **Define success up front, then loop until verified.** State the pass condition before you
+   start; claim done only with the check's output from this session in hand.
+   *Red flag:* "should work", "looks right", "probably passes".
+
+No exceptions for small or obvious tasks — honoring the letter while dodging the intent is a violation.
 
 > Litmus test for any future line in this file: *does it shape how the agent thinks (a behavior), or
 > restate a fact the agent can read from the code?* If it's a fact already in the repo, leave it out.