maxsimcli 5.0.7 → 5.1.0

package/dist/assets/templates/skills/systematic-debugging/SKILL.md

@@ -1,57 +1,66 @@
 ---
 name: systematic-debugging
-description: >-
-  Systematic debugging via reproduce-hypothesize-isolate-verify-fix cycle.
-  Requires evidence at each step. Use when investigating bugs, test failures,
-  unexpected behavior, or runtime errors.
+description: Systematic debugging via reproduce-hypothesize-isolate-verify-fix-confirm cycle. Requires evidence at each step. Use when encountering bugs, test failures, or unexpected behavior.
 ---
 
 # Systematic Debugging
 
 Find the root cause first. Random fixes waste time and create new bugs.
 
-**No fix attempts without understanding root cause.** If you have not completed the REPRODUCE and HYPOTHESIZE steps, you cannot propose a fix.
+**No fix attempts without understanding root cause.** Completing REPRODUCE and HYPOTHESIZE is mandatory before proposing any fix.
 
-## The 5-Step Process
+## The 6-Step Process
 
-### 1. REPRODUCE -- Confirm the Problem
+### 1. REPRODUCE — Confirm the Problem
 
 - Run the failing command or test. Capture the EXACT error output.
-- Can you trigger it reliably? What are the exact steps?
-- If not reproducible: gather more data -- do not guess.
+- Can it be triggered reliably? What are the exact steps?
+- If not reproducible: gather more data — do not guess.
 
-### 2. HYPOTHESIZE -- Form a Theory
+**Output:** Exact reproduction steps and full error output. Nothing moves forward without this.
 
-- Read the error message completely (stack trace, line numbers, exit codes).
+### 2. HYPOTHESIZE — Form a Theory
+
+- Read the error message completely: stack trace, line numbers, exit codes.
 - Check recent changes: `git diff`, recent commits, new dependencies.
 - Trace data flow: where does the bad value originate?
-- State your hypothesis clearly: "I think X is the root cause because Y."
+- State the hypothesis explicitly: "I think X is the root cause because Y."
+
+**Output:** One clear hypothesis with evidence supporting it.
 
-### 3. ISOLATE -- Narrow the Scope
+### 3. ISOLATE — Narrow the Scope
 
 - Find the smallest reproduction case.
 - In multi-component systems, add diagnostic logging at each boundary.
 - Identify which specific layer or component is failing.
 - Compare against working examples in the codebase.
 
-### 4. VERIFY -- Test Your Hypothesis
+**Output:** The smallest failing case and the specific component responsible.
+
+### 4. VERIFY — Test the Hypothesis
+
+- Make the smallest possible change to test the hypothesis.
+- Change one variable at a time — never multiple things simultaneously.
+- If hypothesis is wrong: form a new hypothesis. Do not stack fixes.
 
-- Make the smallest possible change to test your hypothesis.
-- Change one variable at a time -- never multiple things simultaneously.
-- If hypothesis is wrong: form a new hypothesis, do not stack fixes.
+**Output:** Confirmed or rejected hypothesis, with evidence from the test.
 
-### 5. FIX -- Address the Root Cause
+### 5. FIX — Address the Root Cause
 
-- Write a failing test that reproduces the bug.
+- Write a failing test that reproduces the bug (when applicable).
 - Implement a single fix that addresses the root cause.
-- No "while I'm here" improvements -- fix only the identified issue.
+- No "while I'm here" improvements — fix only the identified issue.
 
-### 6. CONFIRM -- Verify the Fix
+**Output:** A minimal fix that targets exactly the identified root cause.
+
+### 6. CONFIRM — Verify the Fix
 
 - Run the original failing test: it must now pass.
 - Run the full test suite: no regressions.
 - Verify the original error no longer occurs.
 
+**Output:** Evidence that the bug is fixed and nothing is broken.
+
 ## Hypothesis Testing Protocol
 
 For each hypothesis:
@@ -59,11 +68,13 @@ For each hypothesis:
 1. **Form:** "I think X is the root cause because Y."
 2. **Design test:** "If X is the cause, then changing Z should produce W."
 3. **Run test:** Execute the change and observe the result.
-4. **Evaluate:** Did the result match the prediction? If yes, proceed to FIX. If no, form a new hypothesis.
+4. **Evaluate:** Did the result match the prediction? If yes, proceed to FIX. If no, discard the hypothesis and form a new one.
+
+Never carry forward a failed hypothesis. Each iteration starts clean.
 
 ## Escalation
 
-If 3+ fix attempts have failed, the issue is likely architectural. Document what you have tried (hypotheses tested, evidence gathered, fixes attempted) and escalate.
+If 3+ fix attempts have failed, the issue is likely architectural. Stop guessing. Document what has been tried — hypotheses tested, evidence gathered, fixes attempted — and escalate or step back to redesign.
 
 ## Common Pitfalls
 
@@ -71,9 +82,10 @@ If 3+ fix attempts have failed, the issue is likely architectural. Document what
 |--------|---------|
 | "I think I know what it is" | Thinking is not evidence. Reproduce first. |
 | "Let me just try this fix" | That is guessing. Complete REPRODUCE and HYPOTHESIZE first. |
-| "Multiple changes at once saves time" | You cannot isolate what worked. You will create new bugs. |
+| "Multiple changes at once saves time" | You cannot isolate what worked. New bugs will appear. |
 | "The issue is simple" | Simple bugs have root causes too. The process is fast for simple bugs. |
+| "The stack trace is enough" | Stack traces show where it crashed, not why. Trace the data. |
 
 Stop immediately if you catch yourself changing code before reproducing, proposing a fix before reading the full error, trying random fixes, or changing multiple things at once.
 
-See also: `/verification-before-completion` for evidence-based confirmation after fixes.
+See also: `verification-before-completion` for evidence-based confirmation after fixes.

package/dist/assets/templates/skills/tdd/SKILL.md

@@ -1,77 +1,74 @@
 ---
 name: tdd
-description: >-
-  Test-driven development with red-green-refactor cycle and atomic commits.
-  Write failing test first, then minimal passing code, then refactor. Use when
-  implementing business logic, API endpoints, data transformations, validation
-  rules, or algorithms.
+description: Enforces red-green-refactor TDD cycle with atomic commits per phase. Use when implementing features, fixing bugs, or when tests should drive the design.
 ---
 
 # Test-Driven Development (TDD)
 
 Write the test first. Watch it fail. Write minimal code to pass. Clean up.
 
-## When to Use TDD
+## When to Use
 
-**Good fit:** Business logic with defined I/O, API endpoints with contracts, data transformations, validation rules, algorithms, state machines.
+**Good fit:** Business logic with defined I/O, API endpoints with contracts, data transformations, validation rules, algorithms, state machines, bug fixes.
 
-**Poor fit:** UI layout, configuration files, build scripts, one-off scripts, mechanical renames.
+**Poor fit:** UI layout, configuration files, build scripts, one-off scripts, mechanical renames, exploratory spikes.
 
 ## The Red-Green-Refactor Cycle
 
-### 1. RED -- Write One Failing Test
+### RED — Write One Failing Test
 
-- Write ONE minimal test describing the desired behavior
-- Test name describes what SHOULD happen, not implementation details
-- Use real code paths -- mocks only when unavoidable (external APIs, databases)
+- Write ONE minimal test describing the desired behavior.
+- Test name describes what SHOULD happen, not implementation details.
+- Use real code paths — mocks only when unavoidable (external APIs, databases, I/O).
 
-### 2. VERIFY RED -- Run the Test
+**Verify RED:** Run the test. It MUST fail with an assertion error — not a syntax error or import failure. If the test passes immediately, it is testing existing behavior. Rewrite it.
 
-- Test MUST fail with an assertion (not error out from syntax or imports)
-- Failure message must match the missing behavior
-- If the test passes immediately, you are testing existing behavior -- rewrite it
+### GREEN — Write Minimal Code
 
-### 3. GREEN -- Write Minimal Code
+- Write the SIMPLEST code that makes the test pass.
+- Do NOT add features the test does not require.
+- Do NOT refactor yet.
 
-- Write the SIMPLEST code that makes the test pass
-- Do NOT add features the test does not require
-- Do NOT refactor yet
+**Verify GREEN:** Run all tests. The new test must pass. ALL existing tests must still pass. If any existing test fails, fix the code — not the tests.
 
-### 4. VERIFY GREEN -- Run All Tests
+### REFACTOR — Clean Up
 
-- The new test MUST pass
-- ALL existing tests MUST still pass
-- If any test fails, fix code -- not tests
+- Remove duplication, improve names, extract helpers.
+- Run tests after every change.
+- Do NOT add new behavior during refactor.
 
-### 5. REFACTOR -- Clean Up (Tests Still Green)
+**Verify REFACTOR:** All tests still pass. No behavior changed.
 
-- Remove duplication, improve names, extract helpers
-- Run tests after every change
-- Do NOT add new behavior during refactor
+### REPEAT
 
-### 6. REPEAT -- Next failing test for next behavior
+Move to the next failing test for the next unit of behavior.
 
 ## Commit Pattern
 
-Each TDD cycle produces 2-3 atomic commits:
+Each TDD cycle produces 2–3 atomic commits:
 
-- **RED commit:** `test({scope}): add failing test for [feature]`
-- **GREEN commit:** `feat({scope}): implement [feature]`
-- **REFACTOR commit (if changes made):** `refactor({scope}): clean up [feature]`
+| Phase | Commit format |
+|-------|---------------|
+| RED | `test({scope}): add failing test for [feature]` |
+| GREEN | `feat({scope}): implement [feature]` |
+| REFACTOR | `refactor({scope}): clean up [feature]` (omit if no changes) |
+
+Keep commits small. One cycle = one feature unit = one commit group.
 
 ## Context Budget
 
-TDD uses approximately 40% more context than direct implementation due to the RED-GREEN-REFACTOR overhead. Plan accordingly for long task lists.
+TDD uses approximately 40% more context than direct implementation due to cycle overhead. Plan accordingly for long task lists.
 
 ## Common Pitfalls
 
 | Excuse | Why It Fails |
 |--------|-------------|
-| "Too simple to test" | Simple code breaks. The test takes 30 seconds. |
-| "I'll add tests after" | Tests written after pass immediately -- they prove nothing. |
+| "Too simple to test" | Simple code breaks. The test takes 30 seconds to write. |
+| "I'll add tests after" | Tests written after pass immediately — they prove nothing. |
 | "I know the code works" | Knowledge is not evidence. A passing test is evidence. |
-| "TDD is slower" | TDD is faster than debugging. Every skip creates debt. |
+| "TDD is slower" | TDD is faster than debugging. Every skipped test creates debt. |
+| "Mocks make this easy" | Over-mocking tests the mock, not the code. Prefer real code paths. |
 
-Stop immediately if you catch yourself writing implementation code before writing a test, writing a test that passes on the first run, skipping the VERIFY RED step, or adding features beyond what the current test requires.
+Stop immediately if you catch yourself writing implementation code before a test, writing a test that passes on the first run, skipping VERIFY RED, or adding features beyond what the current test requires.
 
-See also: `/verification-before-completion` for evidence-based completion claims after TDD cycles.
+See also: `verification-before-completion` for evidence-based completion claims after TDD cycles.

package/dist/assets/templates/skills/using-maxsim/SKILL.md

@@ -1,78 +1,92 @@
 ---
 name: using-maxsim
-description: >-
-  Routes work through MAXSIM's spec-driven workflow: checks planning state,
-  determines active phase, dispatches to the correct MAXSIM command. Use when
-  starting work sessions, resuming work, or choosing which MAXSIM command to run.
+description: Routes work through MaxsimCLI commands based on project state and user intent. Provides command reference and decision routing table. Use when determining which MaxsimCLI command to use or when starting a new session.
 ---
 
-# Using MAXSIM
+# Using MaxsimCLI
 
-MAXSIM is a spec-driven development system. Work flows through phases, plans, and tasks -- not ad-hoc coding.
+MaxsimCLI is a spec-driven development system. Work flows through phases, plans, and tasks — not ad-hoc coding.
 
-**No implementation without a plan.** If there is no `.planning/` directory, run `/maxsim:init` first. If there is no current phase, run `/maxsim:plan` first. If there IS a plan, run `/maxsim:execute` to execute it.
+**No implementation without a plan.** If there is no `.planning/` directory, run `/maxsim:init` first. If there is no current phase, run `/maxsim:plan N` first. If there is a plan, run `/maxsim:execute N` to execute it.
 
-## Routing
+---
+
+## Command Routing Table
+
+Determine user intent, then route to the correct command.
+
+| User Intent | Command |
+|-------------|---------|
+| Start a new project | `/maxsim:init` |
+| Continue where I left off | `/maxsim:go` |
+| Plan next phase | `/maxsim:plan N` |
+| Execute planned work | `/maxsim:execute N` |
+| Fix a bug | `/maxsim:debug` |
+| Quick one-off task | `/maxsim:quick` |
+| Check progress | `/maxsim:progress` |
+| Change settings | `/maxsim:settings` |
+| See all commands | `/maxsim:help` |
+
+---
+
+## Session Start Routing
+
+Before beginning any task in a session:
 
-Before starting any task:
+1. Check for `.planning/` directory — if missing, run `/maxsim:init`
+2. Read `STATE.md` — if a checkpoint exists, resume from it via `/maxsim:go`
+3. Check `ROADMAP.md` — identify the active phase
+4. Route to the correct command using the table above
 
-1. **Check for `.planning/` directory** -- if missing, initialize with `/maxsim:init`
-2. **Check STATE.md** -- resume from last checkpoint if one exists
-3. **Check current phase** -- determine what phase is active in ROADMAP.md
-4. **Route to the correct command** based on the table below
+GitHub Issues with label `maxsim:lesson` or `maxsim:decision` are the source of truth for project learnings and architectural decisions. Read them before planning.
 
-### Command Surface (9 commands)
+---
+
+## Skills Per Agent Type
 
-| Situation | Command |
-|-----------|---------|
-| No `.planning/` directory | `/maxsim:init` |
-| No ROADMAP.md or empty roadmap | `/maxsim:init` |
-| Active phase has no PLAN.md | `/maxsim:plan N` |
-| Active phase has PLAN.md, not started | `/maxsim:execute N` |
-| Phase complete, needs verification | `/maxsim:execute N` (auto-verifies) |
-| Bug found during execution | `/maxsim:debug` |
-| Quick standalone task | `/maxsim:quick` |
-| Check overall status | `/maxsim:progress` |
-| Don't know what to do next | `/maxsim:go` |
-| Change workflow settings | `/maxsim:settings` |
-| Need command reference | `/maxsim:help` |
+Skills load on-demand based on the current task. Each agent type draws from a different set.
 
-### Agent Model (4 agents)
+| Agent | Primary Skills |
+|-------|---------------|
+| Planner | `research`, `brainstorming`, `roadmap-writing`, `project-memory` |
+| Executor | `tdd`, `systematic-debugging`, `verification`, `maxsim-simplify` |
+| Verifier | `verification`, `code-review`, `systematic-debugging` |
+| Researcher | `research`, `project-memory` |
 
-MAXSIM uses 4 generic agent types. Specialization comes from the orchestrator's spawn prompt and on-demand skills, not from separate agent definitions.
+Skills are not auto-loaded. They activate when invoked directly (e.g., `/research`) or when the orchestrator spawns an agent with explicit skill instructions.
+
+---
 
-| Agent | Role | Spawned By |
-|-------|------|-----------|
-| Executor | Implements plans with atomic commits and verified completion | `/maxsim:execute` |
-| Planner | Creates structured PLAN.md files from requirements | `/maxsim:plan` |
-| Researcher | Gathers domain knowledge and codebase context | `/maxsim:plan` (research stage) |
-| Verifier | Reviews code, checks specs, debugs failures | `/maxsim:execute` (review stage), `/maxsim:debug` |
+## GitHub as Source of Truth
 
-### Skills
+All persistent project state lives in GitHub, not in local files that disappear between sessions:
 
-Skills load on-demand based on description matching or direct `/skill-name` invocation. They are not auto-loaded -- each skill activates only when its content is relevant to the current task.
+| Artifact | Location |
+|----------|---------|
+| Phase plans | `.planning/phases/N/PLAN.md` (committed) |
+| Roadmap | `.planning/ROADMAP.md` (committed) |
+| Session state | `.planning/STATE.md` (committed after each checkpoint) |
+| Learnings | GitHub Issues — label `maxsim:lesson` |
+| Decisions | GitHub Issues — label `maxsim:decision` |
 
-| Skill | When It Activates |
-|-------|-------------------|
-| `systematic-debugging` | Investigating bugs, test failures, or unexpected behavior |
-| `tdd` | Implementing business logic, APIs, data transformations |
-| `verification-before-completion` | Claiming work is done, tests pass, builds succeed |
-| `memory-management` | Recurring patterns, errors, or decisions worth persisting |
-| `brainstorming` | Facing architectural choices or design decisions |
-| `roadmap-writing` | Creating or restructuring a project roadmap |
-| `maxsim-simplify` | Reviewing code for duplication, dead code, complexity |
-| `code-review` | Reviewing implementation for security, interfaces, quality |
-| `sdd` | Executing sequential tasks with fresh-agent isolation |
-| `maxsim-batch` | Parallelizing work across independent worktree units |
+---
 
 ## Common Pitfalls
 
-- Writing implementation code without a PLAN.md
-- Skipping `/maxsim:init` because "the project is simple"
-- Ignoring STATE.md checkpoints from previous sessions
+- Writing implementation code without a `PLAN.md`
+- Skipping `/maxsim:init` because the project seems simple
+- Ignoring `STATE.md` checkpoints from previous sessions
 - Working outside the current phase without explicit user approval
-- Making architectural decisions without documenting them in STATE.md
+- Making architectural decisions without recording them as `maxsim:decision` issues
+
+If any of these occur: stop, check the routing table, and follow the workflow.
+
+---
 
-**If any of these occur: stop, check the routing table, follow the workflow.**
+## v6 Changes from v5
 
-See also: `/verification-before-completion` for evidence-based completion claims.
+- `/maxsim:resume-work` replaced by `/maxsim:go`
+- `/maxsim:plan-phase` and `/maxsim:execute-phase` replaced by `/maxsim:plan N` and `/maxsim:execute N`
+- Project memory now uses GitHub Issues instead of local STATE.md comments
+- `research` skill merges former `research-methodology` and `tool-priority-guide`
+- `project-memory` skill replaces `memory-management`

package/dist/assets/templates/skills/verification/SKILL.md

@@ -0,0 +1,167 @@
+---
+name: verification
+description: Evidence-based verification with quality gates, anti-rationalization enforcement, and retry escalation. Merges gate framework, evidence collection, and completion verification into one authoritative source. Use when completing tasks, verifying implementations, or before claiming work is done.
+---
+
+## The Iron Law
+
+No completion claim is valid without fresh verification evidence produced in THIS session. Evidence from a prior session, a prior attempt, or reasoning about what "should" be true does not count. If the evidence was not collected by running a tool call in the current session, it does not exist.
+
+---
+
+## Evidence Block Format
+
+Every claim about task completion, test status, build status, or spec compliance requires an Evidence Block. Produce one per claim.
+
+```
+**CLAIM**: [The specific assertion being made]
+**EVIDENCE**: [Tool name and exact command or action taken]
+**OUTPUT**: [Actual output, quoted verbatim — not paraphrased]
+**VERDICT**: PASS | FAIL | SKIPPED
+```
+
+SKIPPED is only allowed when the claim is explicitly out of scope and the reason is documented. A skipped gate must be acknowledged by the caller.
+
+---
+
+## 4 Quality Gates
+
+Gates run in order. A failure at any gate stops forward progress until resolved.
+
+### Gate 1 — Input Gate
+Run before work begins.
+
+- Spec or task definition exists and is unambiguous
+- Acceptance criteria are stated explicitly
+- Required inputs (files, configs, credentials) are present
+- Scope boundaries are defined — what is in and what is out
+
+Failure action: Return to requester with a clarifying question. Do not guess at requirements.
+
+### Gate 2 — Pre-Action Gate
+Run before executing changes.
+
+- Git state is clean or the working branch is correctly scoped
+- Dependencies are installed and match the lockfile
+- Linter and formatter configs are present
+- No blocking issues from a previous failed run remain in the working tree
+
+Failure action: Resolve the blocking state first. Document what was found and what was done to fix it.
+
+### Gate 3 — Completion Gate
+Run after implementation, before declaring done.
+
+- All tests pass (fresh run, not cached)
+- Build exits with code 0
+- Lint is clean
+- Every acceptance criterion from Gate 1 is addressed with an Evidence Block
+- No files are left in a modified-but-uncommitted state unless that is the intended deliverable
+
+Failure action: Fix failures. Do not skip a failing test. Do not suppress a lint error. Each fix resets the gate — re-run from the top of Gate 3.
+
+### Gate 4 — Quality Gate
+Run after Gate 3 passes.
+
+- Code review concerns (if any were raised) are resolved
+- No regressions introduced — GUARD command confirms this (see below)
+- Evidence Blocks are complete and attached to the work artifact
+- Handoff contract or completion note is written if another agent will consume this output
+
+Failure action: Rework the implementation. If regressions are found, revert before attempting a fix.
+
+---
+
+## What Counts as Evidence
+
+| Claim | Required Evidence | NOT Sufficient |
+|---|---|---|
+| Tests pass | `npm test` output showing pass count and zero failures | "I ran the tests" |
+| Build succeeds | `npm run build` (or equivalent) with exit code 0 shown | "Build should work" |
+| Lint is clean | `npm run lint` output with zero errors and zero warnings | "No obvious lint issues" |
+| File was created | `ls -la <path>` or Read tool output showing the file | "I wrote the file" |
+| Function behaves correctly | Test output or REPL output showing actual return value | "The logic looks right" |
+| API responds correctly | Actual HTTP response body and status code | "The endpoint exists" |
+| Dependency is installed | `package.json` or lockfile entry shown verbatim | "I installed it earlier" |
+| Spec is met | Quoted spec requirement next to quoted output proving it | "This matches the spec" |
+| No regressions | GUARD command output from this session | "Nothing was broken" |
+| Migration ran | Migration log or schema diff output | "I ran the migration" |
+
+---
+
+## Verify + Guard Pattern
+
+Every task execution uses two paired commands.
+
+**VERIFY** — "Did this task accomplish its stated goal?"
+
+Run after implementation. Produces an Evidence Block for each acceptance criterion. If any criterion fails, the task is not done.
+
+**GUARD** — "Did this change break what was already working?"
+
+Run after VERIFY passes. Executes the full test suite and any smoke checks that existed before the task started. If GUARD fails after VERIFY passes, the implementation introduced a regression.
+
+### Regression Protocol
+
+1. VERIFY passes, GUARD fails: attempt rework, limit 2 rework cycles
+2. After 2 rework cycles: revert the change entirely, escalate to user
+3. Do not merge a change where GUARD is failing
+
+---
+
+## Anti-Rationalization Table
+
+These phrases indicate a verification failure. They are never acceptable as evidence.
+
+| Forbidden Phrase | Why It Fails |
+|---|---|
+| "should work" | Describes expectation, not observed outcome |
+| "I already checked" | Not verifiable in this session |
+| "tests were passing before" | Stale evidence; fresh run required |
+| "this is obviously correct" | Correctness is measured, not assessed by inspection |
+| "I think it's fine" | No tool output, no claim |
+| "the logic is sound" | Logic can be sound and still produce wrong output |
+| "nothing changed in that area" | Changes in dependencies, configs, and imports are invisible to this claim |
+| "it worked in my local run" | Local run is not this session's evidence unless tool output is shown |
+| "we can verify later" | Verification deferred is verification skipped |
+| "this is low risk" | Risk level does not substitute for evidence |
+
+---
+
+## Retry Logic
+
+### Attempt Counting
+
+Each task starts at attempt 1. A failed gate that triggers a rework cycle increments the attempt counter. Attempt count resets only when the task scope changes materially.
+
+### Per-Attempt Rules
+
+- **Attempt 1**: Execute normally. Collect full evidence. If gates pass, complete.
+- **Attempt 2**: Fresh context. Do not carry forward assumptions from attempt 1. Re-read the spec. Re-run all gates from Gate 1.
+- **Attempt 3**: Fresh agent context. Treat this as a cold start. Diagnose why attempts 1 and 2 failed before touching any code.
+
+### After 3 Failures
+
+Escalate to the user. The escalation must include:
+
+1. The original task spec (quoted)
+2. What was attempted in each of the 3 runs (brief, factual)
+3. The specific gate that failed each time and the exact error output
+4. A diagnostic summary: is this a spec problem, an environment problem, or an implementation problem?
+5. A proposed next step (rewrite spec, fix environment, reduce scope)
+
+Do not attempt a 4th run without user acknowledgment and revised instructions.
+
+---
+
+## Common Pitfalls
+
+| Pitfall | Symptom | Correct Behavior |
+|---|---|---|
+| Caching test results | Reporting pass without re-running | Always run tests fresh; use `--no-cache` or equivalent |
+| Partial lint scope | Running lint on one file, claiming lint is clean | Run lint on the entire affected module or project |
+| Missing Gate 1 | Starting work before spec is confirmed | Always confirm acceptance criteria exist before writing code |
+| Evidence copied from prior session | Referencing output not produced in this session | All evidence must come from tool calls in the current session |
+| Verifying only the happy path | Tests pass but edge cases are untested | GUARD must include regression tests, not only new tests |
+| Skipping Gate 4 after Gate 3 passes | Declaring done without regression check | Gate 3 and Gate 4 are both required; neither is optional |
+| Conflating "no errors" with "correct output" | Exit code 0 but wrong behavior | Evidence must show correct output, not just absence of error |
+| Writing evidence after the fact | Constructing output from memory | Run the command, capture the output, paste it verbatim |