Repository navigation
feat(wiz-auto): enforce absolute autonomy - never stop for confirmation - #4
Conversation
|
Caution Review failedThe pull request is closed. 📝 WalkthroughWalkthroughA single prompt configuration file ( Changes
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~18 minutes
Pre-merge checks and finishing touches✅ Passed checks (3 passed)
📜 Recent review detailsConfiguration used: CodeRabbit UI Review profile: CHILL Plan: Pro 📒 Files selected for processing (1)
Comment |
6fb9faa to
9f379bb
Compare
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 1
🧹 Nitpick comments (3)
.cursor/commands/wiz-auto.md (3)
51-61: Consolidate or clarify purpose of redundant "CRITICAL BEHAVIOR" section.Lines 51-61 echo the FORBIDDEN BEHAVIORS section (lines 22-36) using affirmative language instead of prohibitive. While repetition reinforces the behavior, this creates document bloat. Either:
- Consolidate: Merge into a single section with both negative ("Don't ask...") and positive ("Do execute...") frames, OR
- Clarify Purpose: If repetition is intentional for agent training via redundancy, add a comment explaining pedagogical intent.
This refactoring is optional but would improve maintainability.
1571-1641: Redundancy: Forbidden phrases listed 4 separate times across document.The forbidden outputs appear in:
- Lines 24-35: FORBIDDEN BEHAVIORS (15 items)
- Lines 113-121: "IF ANALYST RETURNS PROCEED" (7 items)
- Lines 1572-1576: "FINAL REMINDER" (5 items)
- Lines 1848-1857: Code block "OTHER WRONG OUTPUTS" (10 items)
While repetition may be intentional for agent reinforcement, this makes maintenance difficult. Any future update to forbidden phrases requires changes in 4+ locations.
Recommendation: Define forbidden phrases once (e.g., in a clearly marked section), then reference that section. Example:
## Forbidden Phrases (Master List) - [centralized list of all forbidden outputs] ## Section A [Reference: See Forbidden Phrases list above] ## Section B [Reference: See Forbidden Phrases list above]
1802-1872: Optional: Tighten verbose example section and fix repetitive word structure.The "Example 2: WRONG" section (lines 1802-1843) effectively demonstrates incorrect behavior but is verbose. The bullets at lines 1836-1843 mostly restate lines 105-141. Suggestion: Condense to 2-3 bullets and cross-reference the earlier "EVERYTHING ELSE = KEEP GOING" rule.
Additionally, line 1882 has three sentences beginning with "NO":
NO OTHER STOPS. NO QUESTIONS. NO CONFIRMATIONS.While emphatic, this violates style guidelines. Rewrite as:
There are no other stops, no questions, and no confirmations needed.These are minor readability improvements.
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (1)
.cursor/commands/wiz-auto.md(7 hunks)
🧰 Additional context used
🪛 LanguageTool
.cursor/commands/wiz-auto.md
[style] ~30-~30: Consider using polite language here.
Context: ...* - JUST CONTINUE NOW! - ❌ NEVER say "Let me know if you'd like me to continue" - CONTI...
(INSERT_PLEASE)
[style] ~115-~115: Consider using polite language here.
Context: ...milestones are available" - DO NOT say "Let me know when you want to continue" - DO NOT say "Sho...
(INSERT_PLEASE)
[style] ~1882-~1882: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...HAT'S IT. NO OTHER STOPS. NO QUESTIONS. NO CONFIRMATIONS. JUST KEEP GOING.** ### ...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
🔇 Additional comments (2)
.cursor/commands/wiz-auto.md (2)
100-124: Clarify handling of agent-level errors within absolute loop mandate.Line 105 states "EVERYTHING ELSE = KEEP GOING," but this is too absolute. Step 10 (Error Handling, line 1686) mentions "error and halt" scenarios. These appear contradictory:
- Absolute autonomy rule (line 105): Treat all non-HALT/non-complete states as "continue"
- Error handling (line 1686): "Analyst decision unclear: Error and halt"
What is the precedence if:
- Specialist review fails 5 times in a row (Step 7.3)?
- Analyst returns malformed output?
- Git commit fails unexpectedly?
Suggestion: Add explicit error boundaries after line 123. Example:
- DO NOT continue if same error repeats >3 times - DO escalate network/file-system errors immediately - DO halt if specialist review loops indefinitely
1921-1935: Reconsider framing of trust in final reminder—may overemphasize speed over safety.Lines 1921-1935 appeal to "honoring user trust" by executing indefinitely without interruption. However, this frames the user's motivation as purely about speed ("user's time is valuable. They don't want to babysit you.").
Alternative framing: Users may value autonomy AND safety equally. A user who runs
/wiz-autotrusts the system to:
- Execute continuously when safe, AND
- Halt immediately when decisions are needed
Suggestion: Reframe lines 1927-1934 to emphasize both:
They trust you to maintain momentum while catching issues that need human judgment. HONOR THAT TRUST: Execute fast AND halt early when uncertain.This acknowledges the autonomy requirement while leaving room for safety-first decision making.
| --- | ||
|
|
||
| ## 🚨🚨🚨 ABSOLUTE AUTONOMY REQUIREMENT 🚨🚨🚨 | ||
|
|
||
| **YOU ARE FULLY AUTONOMOUS. NEVER STOP UNLESS:** | ||
| 1. **ALL milestones are COMPLETE** → Celebrate with 🎉 | ||
| 2. **Analyst explicitly returns HALT** with specific questions needing human decisions | ||
|
|
||
| **THERE ARE NO OTHER VALID REASONS TO STOP.** | ||
|
|
||
| ### ❌ FORBIDDEN BEHAVIORS - NEVER DO THESE: | ||
|
|
||
| - ❌ **NEVER ask "Do you want me to continue?"** - YES, ALWAYS CONTINUE! | ||
| - ❌ **NEVER ask "Should I proceed?"** - YES, ALWAYS PROCEED! | ||
| - ❌ **NEVER ask "Would you like a context refresh?"** - NO, NEVER! | ||
| - ❌ **NEVER ask "Should I review the progress?"** - NO, JUST KEEP GOING! | ||
| - ❌ **NEVER ask "Do you want to review the changes?"** - NO! | ||
| - ❌ **NEVER say "Ready to continue when you are"** - JUST CONTINUE NOW! | ||
| - ❌ **NEVER say "Let me know if you'd like me to continue"** - CONTINUE IMMEDIATELY! | ||
| - ❌ **NEVER pause for "user confirmation"** - YOU HAVE FULL AUTHORITY! | ||
| - ❌ **NEVER stop to "summarize progress"** mid-execution | ||
| - ❌ **NEVER offer options or choices** unless analyst says HALT | ||
| - ❌ **NEVER wait for user input** between milestones | ||
| - ❌ **NEVER stop because "it's been a while"** or "we've done several milestones" | ||
|
|
||
| ### ✅ CORRECT BEHAVIOR: | ||
|
|
||
| - ✅ **Execute milestone → Commit → Analyst check → If PROCEED → Execute next milestone** | ||
| - ✅ **Keep going automatically** - no pauses, no confirmations, no questions | ||
| - ✅ **Only stop for HALT** when analyst has specific questions requiring human decisions | ||
| - ✅ **Complete the entire project** if possible in one session | ||
|
|
||
| **THE USER NEVER NEEDS A CONTEXT REFRESH.** | ||
| **THE USER NEVER NEEDS TO REVIEW PROGRESS.** | ||
| **THE USER NEVER NEEDS TO CONFIRM CONTINUATION.** | ||
| **JUST. KEEP. GOING. FOREVER.** | ||
|
|
||
| --- |
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
# First, get the file size to understand its scope
wc -l .cursor/commands/wiz-auto.mdRepository: NSXBet/wiz-cursor
Length of output: 94
🏁 Script executed:
#!/bin/bash
# Read the ABSOLUTE AUTONOMY section (lines 1-100 to get context)
sed -n '1,100p' .cursor/commands/wiz-auto.mdRepository: NSXBet/wiz-cursor
Length of output: 4685
🏁 Script executed:
#!/bin/bash
# Read the error handling section cited in the review (lines 1676-1700)
sed -n '1676,1700p' .cursor/commands/wiz-auto.mdRepository: NSXBet/wiz-cursor
Length of output: 1101
🏁 Script executed:
#!/bin/bash
# Search the entire file for resource limits, timeouts, or execution boundaries
rg -n -i 'timeout|resource|limit|circuit|max.*milestone|max.*execution|boundary|fail.*safe' .cursor/commands/wiz-auto.mdRepository: NSXBet/wiz-cursor
Length of output: 43
🏁 Script executed:
#!/bin/bash
# Search for analyst/specialist sections and any safeguards
rg -n -C3 'analyst.*malfunction|analyst.*fail|specialist.*loop|specialist.*retry' .cursor/commands/wiz-auto.mdRepository: NSXBet/wiz-cursor
Length of output: 458
🏁 Script executed:
#!/bin/bash
# Look for the actual loop implementation and specialist review logic
sed -n '1066,1150p' .cursor/commands/wiz-auto.mdRepository: NSXBet/wiz-cursor
Length of output: 3047
🏁 Script executed:
#!/bin/bash
# Search for specialist review implementation and retry logic
rg -n -B2 -A5 'specialist.*review|Step 7|Step 8|wiz_specialist_review' .cursor/commands/wiz-auto.md | head -60Repository: NSXBet/wiz-cursor
Length of output: 2657
🏁 Script executed:
#!/bin/bash
# Read the full Step 7.3 logic to see if there's a max retry
sed -n '1414,1500p' .cursor/commands/wiz-auto.mdRepository: NSXBet/wiz-cursor
Length of output: 2931
🏁 Script executed:
#!/bin/bash
# Look at Step 9 (analyst invocation) and check for safeguards
sed -n '1500,1600p' .cursor/commands/wiz-auto.mdRepository: NSXBet/wiz-cursor
Length of output: 3518
🏁 Script executed:
#!/bin/bash
# Read full Step 10 to see HALT parsing and any analyst safeguards
sed -n '1600,1700p' .cursor/commands/wiz-auto.mdRepository: NSXBet/wiz-cursor
Length of output: 4023
Based on my verification of the .cursor/commands/wiz-auto.md file, I can now provide the rewritten review comment.
Verify and document safeguards for unbounded auto-execution loops.
The "ABSOLUTE AUTONOMY REQUIREMENT" section correctly enforces loop continuation (lines 14-47), but the implementation lacks documented execution boundaries. Verified findings:
-
No max milestone limit: The loop counter at line 1068 (
MILESTONES_COMPLETED=0) is incremented for logging only; no condition prevents executing 50+ milestones in a single session. -
No execution timeout: No mention of max execution time, max execution hours, or wall-clock limits.
-
Specialist review loop without max retry (lines 1414–1427): The loop "repeats until ALL specialists approve" with no iteration cap. If issues persist or reoccur, this could loop indefinitely. Error handling (line 1691) mentions "fails repeatedly" but defines no threshold.
-
Analyst malfunction path: No safeguard if analyst consistently returns PROCEED (e.g., max 5 consecutive PROCEEDs before requiring human review).
-
Informed consent ambiguity (lines 8–10): The entry warning states this will "execute multiple milestones (potentially dozens)" but provides no documented upper bounds, making it unclear what "unbounded" means to the user invoking
/wiz-auto.
Recommended additions (after line 20):
- Document max milestones per run (e.g., 20–50)
- Define execution timeout (e.g., 2–4 hours)
- Specify max specialist review retries before human halt (e.g., 5)
- Add fallback: if analyst returns PROCEED > N times, require explicit user confirmation for continuation
These boundaries preserve autonomy while preventing runaway execution and clarifying user expectations.
🧰 Tools
🪛 LanguageTool
[style] ~30-~30: Consider using polite language here.
Context: ...* - JUST CONTINUE NOW! - ❌ NEVER say "Let me know if you'd like me to continue" - CONTI...
(INSERT_PLEASE)
🤖 Prompt for AI Agents
In .cursor/commands/wiz-auto.md around lines 12 to 49, the "ABSOLUTE AUTONOMY
REQUIREMENT" section enforces continuous execution but lacks documented
safeguards against unbounded loops; add concise policy language right after line
20 that defines concrete execution limits: set a configurable max milestones per
run (e.g., default 25, configurable 20–50), a wall-clock execution timeout
(e.g., default 3 hours), a max specialist review retry count (e.g., 5 retries)
and a max consecutive analyst PROCEEDs (e.g., 5) after which the system must
require explicit human confirmation; also add a short note that these defaults
are configurable and where to change them (config file or flags).
There was a problem hiding this comment.
Actionable comments posted: 3
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (3)
.cursor/commands/wiz-phases.md (1)
467-513: Escape sed replacement values in wiz_render_template to avoid corrupt output or injection.Current substitution does not escape
&and backslashes in values, which sed treats specially, breaking rendering for common inputs (paths, JSON, URLs).Apply escaping before substitution:
wiz_render_template() { @@ - local rendered="$template" + local rendered="$template" @@ - local var_value="${var_assignment#*=}" + local var_value="${var_assignment#*=}" + # Escape sed replacement specials: \ and & + local esc_value + esc_value="$(printf '%s' "$var_value" | sed -e 's/[\\/&]/\\&/g')" @@ - rendered=$(echo "$rendered" | sed "s|{{${var_name}}}|${var_value}|g") + rendered=$(printf '%s' "$rendered" | sed "s|{{${var_name}}}|${esc_value}|g") done.cursor/commands/wiz-auto.md (1)
1309-1315: Do not include secrets/config files in specialist diff.The guidance urges sending “ALL config files” and current exclude list misses common secret paths (.env, keys, certs). Risk of credential leakage.
Harden the diff generation:
CHANGES_DIFF=$(git diff HEAD -- . \ ':(exclude)*.jpg' ':(exclude)*.jpeg' ':(exclude)*.png' ':(exclude)*.gif' \ ':(exclude)*.pdf' ':(exclude)*.zip' ':(exclude)*.tar' ':(exclude)*.gz' \ ':(exclude)*.exe' ':(exclude)*.dll' ':(exclude)*.so' ':(exclude)*.dylib' \ - ':(exclude)*.db' ':(exclude)*.sqlite' ':(exclude)*.csv' ':(exclude)*.parquet') + ':(exclude)*.db' ':(exclude)*.sqlite' ':(exclude)*.csv' ':(exclude)*.parquet' \ + ':(exclude).env' ':(exclude).env.*' ':(exclude)*.env' \ + ':(exclude)*secret*' ':(exclude)*secrets*' \ + ':(exclude)*id_rsa*' ':(exclude)*.pem' ':(exclude)*.key' ':(exclude)*.p12' \ + ':(exclude)*.keystore' ':(exclude)*.pfx')Also document that secrets should be redacted when present in diffs.
.cursor/commands/wiz-review-phase.md (1)
169-201: Fix command naming: change all/wiz:forms to/wiz-to match canonical repository convention.The repository consistently uses hyphen-based command names (
/wiz-review-phase,/wiz-phases,/wiz-prd) throughout documentation, tests, and all other files. The code at lines 169-201 incorrectly uses the colon form (/wiz:review-phase,/wiz:phases,/wiz:prd). Update all three instances in the usage string and error messages to use hyphens.
♻️ Duplicate comments (1)
.cursor/commands/wiz-auto.md (1)
1416-1431: Cap specialist review retries and add backoff to avoid livelocks.Looping “until ALL specialists approve” can spin forever on non‑actionable or conflicting feedback. Add a retry cap (e.g., 3–5), exponential backoff, and escalate to HALT with a consolidated report when exceeded.
🧹 Nitpick comments (7)
.cursor/commands/wiz-help.md (1)
162-168: Consider broadening the flag pattern to improve coverage.The grep pattern
^\s*-\-[a-z-]+.*:matches flags in the format--flag-name:but may miss flags with uppercase letters, numbers, or underscores (e.g.,--myFlag,--MY_FLAG,--flag_name). If the codebase uses diverse naming conventions for flags, this pattern could silently skip valid options.If comprehensive coverage is desired, consider:
grep -E '^\s*-{1,2}[A-Za-z0-9_-]+.*:' "$command_file" | head -5Otherwise, document the naming convention assumption if the current pattern is intentional.
.cursor/commands/wiz-phases.md (1)
511-513: Guard optional tooling (jq) with early checks to fail fast, not mid‑pipeline.Multiple helpers assume jq exists; add a single preflight check with a clear error to improve UX.
Example:
+_wiz_require_tools() { + for t in jq awk sed find; do + command -v "$t" >/dev/null 2>&1 || { echo "ERROR: Missing required tool: $t" >&2; return 1; } + done +} + # Step 1 set -euo pipefail +_wiz_require_tools || exit 1Also applies to: 146-147, 232-252
.cursor/commands/wiz-auto.md (1)
1354-1370: Single‑message, all‑specialists review can exceed context limits; chunk by language/file sets.Large diffs across multiple stacks easily blow token limits. Send per‑language chunks with concise summaries and allow specialists to request more context.
Suggested policy:
- Split CHANGES_DIFF by language.
- For each specialist: send only relevant chunk + file list; provide a “more context” on demand.
- Enforce a max payload size; fallback to batched reviews.
.cursor/commands/wiz-review-phase.md (1)
260-271: Capitalize “Markdown” when referring to the format.Minor editorial polish for consistency.
Also applies to: 415-416
.cursor/agents/wiz-planner.md (1)
397-401: Avoid informal “gonna” in formal docs.Replace with “are not going to” to keep tone professional.
.cursor/commands/wiz-review-milestone.md (1)
285-285: Capitalize "Markdown" as proper noun.Lines 285 and 293 reference "markdown code blocks" and "markdown in code blocks." Per English language conventions, "Markdown" is a proper noun and should be capitalized consistently.
Apply this diff:
- The `wiz-reviewer` agent **returns audit report content** as markdown code blocks in its response + The `wiz-reviewer` agent **returns audit report content** as Markdown code blocks in its response- 1. Agent returns audit report as markdown in code blocks + 1. Agent returns audit report as Markdown in code blocksAlso applies to: 293-293
.cursor/commands/wiz-milestones.md (1)
289-362: Verify YAML frontmatter parsing is robust to edge cases.The context loading functions (lines 289–362) parse YAML frontmatter using sed/grep/jq chains. This approach is brittle and may fail on:
- Frontmatter containing
---(common in YAML content)- Quoted or escaped values in frontmatter (e.g.,
description: "foo: bar")- YAML syntax variations (e.g., trailing colons, multiline values)
Since these functions are embedded in documentation for user/agent execution, edge cases in malformed files could cause silent failures or incorrect parsing.
Consider:
- Adding validation to detect if frontmatter is well-formed before parsing.
- Using a more robust YAML parser (e.g., Python with PyYAML) if edge cases are likely.
- Documenting limitations of the current parser for frontmatter format constraints.
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (22)
.cursor/agents/wiz-csharp-specialist.md(16 hunks).cursor/agents/wiz-docker-specialist.md(23 hunks).cursor/agents/wiz-go-specialist.md(27 hunks).cursor/agents/wiz-java-specialist.md(15 hunks).cursor/agents/wiz-milestone-analyst.md(7 hunks).cursor/agents/wiz-planner.md(22 hunks).cursor/agents/wiz-python-specialist.md(17 hunks).cursor/agents/wiz-reviewer.md(31 hunks).cursor/agents/wiz-typescript-specialist.md(15 hunks).cursor/commands/wiz-auto.md(13 hunks).cursor/commands/wiz-help.md(8 hunks).cursor/commands/wiz-milestones.md(25 hunks).cursor/commands/wiz-next.md(20 hunks).cursor/commands/wiz-phases.md(11 hunks).cursor/commands/wiz-resume.md(3 hunks).cursor/commands/wiz-review-milestone.md(6 hunks).cursor/commands/wiz-review-phase.md(5 hunks).cursor/commands/wiz-status-simple.md(0 hunks).cursor/commands/wiz-status.md(1 hunks).cursor/commands/wiz-validate-all.md(2 hunks)docs/agents.md(1 hunks)docs/commands.md(2 hunks)
💤 Files with no reviewable changes (1)
- .cursor/commands/wiz-status-simple.md
✅ Files skipped from review due to trivial changes (10)
- .cursor/agents/wiz-csharp-specialist.md
- .cursor/commands/wiz-resume.md
- .cursor/agents/wiz-typescript-specialist.md
- docs/commands.md
- .cursor/agents/wiz-python-specialist.md
- .cursor/commands/wiz-next.md
- .cursor/commands/wiz-validate-all.md
- docs/agents.md
- .cursor/commands/wiz-status.md
- .cursor/agents/wiz-docker-specialist.md
🧰 Additional context used
🪛 LanguageTool
.cursor/agents/wiz-milestone-analyst.md
[style] ~307-~307: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...oaches? - Are requirements ambiguous? - Are there undefined terms or assumptions? -...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
.cursor/agents/wiz-go-specialist.md
[grammar] ~527-~527: Ensure spelling is correct
Context: ...ith many checks enabled by default ### gofmt bash gofmt -w . Standard formatting - always run before ...
(QB_NEW_EN_ORTHOGRAPHY_ERROR_IDS_1)
.cursor/commands/wiz-review-phase.md
[uncategorized] ~260-~260: Did you mean the formatting language “Markdown” (= proper noun)?
Context: ...nt returns review report content as markdown code blocks in its response - The main ...
(MARKDOWN_NNP)
[uncategorized] ~268-~268: Did you mean the formatting language “Markdown” (= proper noun)?
Context: ...below 1. Agent returns review report as markdown in code blocks 1. Main agent writes the...
(MARKDOWN_NNP)
[uncategorized] ~415-~415: Did you mean the formatting language “Markdown” (= proper noun)?
Context: ... Return the complete review report as a markdown code block. The main agent will write i...
(MARKDOWN_NNP)
.cursor/commands/wiz-review-milestone.md
[uncategorized] ~285-~285: Did you mean the formatting language “Markdown” (= proper noun)?
Context: ...ent returns audit report content as markdown code blocks in its response - The main ...
(MARKDOWN_NNP)
[uncategorized] ~293-~293: Did you mean the formatting language “Markdown” (= proper noun)?
Context: ... below 1. Agent returns audit report as markdown in code blocks 1. Main agent writes the...
(MARKDOWN_NNP)
.cursor/commands/wiz-auto.md
[style] ~31-~31: Consider using polite language here.
Context: ...* - JUST CONTINUE NOW! - ❌ NEVER say "Let me know if you'd like me to continue" - CONTI...
(INSERT_PLEASE)
[style] ~116-~116: Consider using polite language here.
Context: ...milestones are available" - DO NOT say "Let me know when you want to continue" - DO NOT say "Sho...
(INSERT_PLEASE)
[style] ~1060-~1060: Consider using a different verb for a more formal wording.
Context: ...ist Review** ← Specialist reviews diff, fix issues if found 1. Create Commit ← ...
(FIX_RESOLVE)
[style] ~1885-~1885: Three successive sentences begin with the same word. Consider rewording the sentence or use a thesaurus to find a synonym.
Context: ...HAT'S IT. NO OTHER STOPS. NO QUESTIONS. NO CONFIRMATIONS. JUST KEEP GOING.** ### ...
(ENGLISH_WORD_REPEAT_BEGINNING_RULE)
.cursor/agents/wiz-planner.md
[style] ~400-~400: The word ‘gonna’ is informal.
Context: ...ty only when needed - YAGNI (You Aren't Gonna Need It) Under-specifying: - "Add...
(GONNA)
.cursor/agents/wiz-reviewer.md
[uncategorized] ~685-~685: If this is a compound adjective that modifies the following noun, use a hyphen.
Context: ...critical vulnerabilities - Dependencies up to date - Transitive dependencies checked **Fi...
(EN_COMPOUND_ADJECTIVE_INTERNAL)
🔇 Additional comments (3)
.cursor/commands/wiz-help.md (2)
91-102: Clarify the four-backtick code fence syntax.Line 91 uses four backticks (````bash) instead of the standard three. Verify whether this is intentional (e.g., to support nested code blocks or a custom parser) or a formatting artifact. Standard markdown and most renderers expect three backticks.
40-44: Help documentation expansion is well-structured.The additions of "Common options if available" and "Troubleshooting tips" to the command-specific help output enhance usability without introducing complexity. The implementation cleanly integrates these new sections into the generation logic.
.cursor/commands/wiz-milestones.md (1)
10-20: Clarify execution strategy: Is delegation prohibited or conditional?Lines 10–20 state emphatically: "YOU MUST WRITE FILES TO DISK YOURSELF. Do not delegate to subagents for file operations." However, lines 547–701 describe "parallel processing" using Task tool delegation to wiz-planner agent.
These statements appear contradictory. Confirm whether:
- The no-delegation rule applies universally, and "parallel processing" uses a different mechanism, OR
- Delegation is permitted for Option A (parallel), but Option B (sequential) requires direct file writing, OR
- One of these sections reflects outdated guidance and should be removed/clarified.
Clarify the execution model in lines 10–20 and lines 547–701 to ensure users understand when delegation is permitted vs. prohibited.
Also applies to: 547-701
| ## 🚨🚨🚨 ABSOLUTE AUTONOMY REQUIREMENT 🚨🚨🚨 | ||
|
|
||
| **YOU ARE FULLY AUTONOMOUS. NEVER STOP UNLESS:** | ||
|
|
||
| 1. **ALL milestones are COMPLETE** → Celebrate with 🎉 | ||
| 1. **Analyst explicitly returns HALT** with specific questions needing human decisions | ||
|
|
||
| **THERE ARE NO OTHER VALID REASONS TO STOP.** | ||
|
|
||
| ### ❌ FORBIDDEN BEHAVIORS - NEVER DO THESE: | ||
|
|
||
| - ❌ **NEVER ask "Do you want me to continue?"** - YES, ALWAYS CONTINUE! | ||
| - ❌ **NEVER ask "Should I proceed?"** - YES, ALWAYS PROCEED! | ||
| - ❌ **NEVER ask "Would you like a context refresh?"** - NO, NEVER! | ||
| - ❌ **NEVER ask "Should I review the progress?"** - NO, JUST KEEP GOING! | ||
| - ❌ **NEVER ask "Do you want to review the changes?"** - NO! | ||
| - ❌ **NEVER say "Ready to continue when you are"** - JUST CONTINUE NOW! | ||
| - ❌ **NEVER say "Let me know if you'd like me to continue"** - CONTINUE IMMEDIATELY! | ||
| - ❌ **NEVER pause for "user confirmation"** - YOU HAVE FULL AUTHORITY! | ||
| - ❌ **NEVER stop to "summarize progress"** mid-execution | ||
| - ❌ **NEVER offer options or choices** unless analyst says HALT | ||
| - ❌ **NEVER wait for user input** between milestones | ||
| - ❌ **NEVER stop because "it's been a while"** or "we've done several milestones" | ||
|
|
||
| ### ✅ CORRECT BEHAVIOR: | ||
|
|
||
| - ✅ **Execute milestone → Commit → Analyst check → If PROCEED → Execute next milestone** | ||
| - ✅ **Keep going automatically** - no pauses, no confirmations, no questions | ||
| - ✅ **Only stop for HALT** when analyst has specific questions requiring human decisions | ||
| - ✅ **Complete the entire project** if possible in one session | ||
|
|
||
| **THE USER NEVER NEEDS A CONTEXT REFRESH.** | ||
| **THE USER NEVER NEEDS TO REVIEW PROGRESS.** | ||
| **THE USER NEVER NEEDS TO CONFIRM CONTINUATION.** | ||
| **JUST. KEEP. GOING. FOREVER.** | ||
|
|
There was a problem hiding this comment.
Resume prompt contradicts “Absolute Autonomy”; auto‑resume instead of pausing.
Section mandates “NEVER stop unless HALT or complete”, but Step 2 prints an interactive choice and exits. Make the command auto‑resume the in‑progress milestone without user input.
Suggested change (remove prompt + exit, log and proceed):
if [[ -f "$RESUME_STATE_FILE" ]]; then
RESUME_STATUS=$(jq -r '.status // "unknown"' "$RESUME_STATE_FILE")
if [[ "$RESUME_STATUS" == "in_progress" ]]; then
- RESUME_ID=$(jq -r '.milestone_id' "$RESUME_STATE_FILE")
- RESUME_STARTED=$(jq -r '.started_at' "$RESUME_STATE_FILE")
- wiz_log_info "Found in-progress milestone: $RESUME_ID"
- echo ""
- echo "⚠️ In-Progress Milestone Found"
- ...
- echo "Or delete $RESUME_STATE_FILE to skip to next"
- exit 0
+ RESUME_ID=$(jq -r '.milestone_id' "$RESUME_STATE_FILE")
+ wiz_log_info "Auto-resuming in-progress milestone: $RESUME_ID"
+ # Continue without prompting; loop will pick up resume state
fi
fiAlso applies to: 1007-1033
🧰 Tools
🪛 LanguageTool
[style] ~31-~31: Consider using polite language here.
Context: ...* - JUST CONTINUE NOW! - ❌ NEVER say "Let me know if you'd like me to continue" - CONTI...
(INSERT_PLEASE)
🤖 Prompt for AI Agents
In .cursor/commands/wiz-auto.md around lines 14 to 49 (and similarly for the
duplicate block at 1007–1033), the resume step contradicts the "ABSOLUTE
AUTONOMY" policy by presenting an interactive prompt and exiting; change the
behavior to remove the prompt and exit path, instead auto-resume the in‑progress
milestone: on reaching the resume step log a concise informational message that
you are auto‑resuming, restore any necessary in‑memory/state pointers for the
milestone, and continue execution immediately without waiting for user input or
terminating the process.
| **THERE ARE ONLY 2 VALID REASONS TO STOP:** | ||
|
|
||
| 1. No more TODO milestones found (Step 4) → Celebrate with big 🎉 message | ||
| 1. Analyst returns HALT decision with SPECIFIC QUESTIONS (Step 10) → Show questions and exit | ||
|
|
||
| 1. No more TODO milestones found (Step 4) → Celebrate with big message | ||
| 2. Analyst returns HALT decision (Step 10) → Show questions and exit | ||
| **EVERYTHING ELSE = KEEP GOING** | ||
|
|
There was a problem hiding this comment.
Stop‑condition inconsistency: doc says 2 reasons, later lists 3. Align the contract.
Either explicitly allow “error abort” as an exceptional stop or keep the “two reasons” wording consistent across both sections.
Also applies to: 1658-1663
🤖 Prompt for AI Agents
In .cursor/commands/wiz-auto.md around lines 101 to 107 (and also adjust lines
1658 to 1663), the doc claims “THERE ARE ONLY 2 VALID REASONS TO STOP” but later
describes three possible stop conditions; reconcile this by either (A) updating
the wording to explicitly allow a third “error abort/exceptional stop” reason
and add a brief explanatory sentence listing the accepted exceptional stop, or
(B) remove the extra stop from the later list so it matches the “two reasons”
contract; apply the same consistent wording and list change to both locations.
| echo "✓ Appended milestone" | ||
| } | ||
| ``` | ||
| ```` |
There was a problem hiding this comment.
Fix markdown syntax error: mismatched backtick count in code fence.
Line 642 closes a code block with five backticks (`````) instead of three, breaking markdown rendering. This fence should close the code block opened at line 599.
Apply this diff:
- ````
+ ```Also applies to: 701-701
🤖 Prompt for AI Agents
In .cursor/commands/wiz-milestones.md around lines 599–642 and 701, there are
mismatched Markdown code fences: line 642 currently closes a code block with
five backticks (`````) and line 701 has a similar issue; replace each incorrect
five-backtick closing fence with the proper three-backtick fence (```) so the
code blocks opened at line 599 and the corresponding block near 701 are
correctly closed, preserving surrounding whitespace and ensuring no other fence
markers are altered.
9f379bb to
3ec7451
Compare
1 participant
The agent should now understand that running /wiz-auto grants full
permission to execute indefinitely without checking in with the user.
Summary by CodeRabbit
✏️ Tip: You can customize this high-level summary in your review settings.