SECURITY REVIEW
Not yet assessed
Review the original instructions and requested permissions before installing.
No security review is available for this catalog entry yet.
Community / AGENT SKILL
direct-implementation
0
Rapid single-turn direct implementation workflow for unambiguous, low-risk, single-file code changes and bug fixes with known root causes. Enforces pre-action checks and post-edit sanity verification.
Understand the trade-offs.
SKILL QUALITY
Not yet assessed
How clearly the skill guides your agent, how complete its workflow is, and how you can check the outcome.
No quality assessment is available for this catalog entry yet.
The full skill.
Original instructions from the publisher’s SKILL.md
<!-- Generated from workflows/direct-implementation.md by tools/generate_integrations.py. Edit the source file, not this one. This is an execution WORKFLOW packaged as an Agent Skill so it is discoverable and directly invocable ("/direct-implementation") in every compatible tool. -->
# Workflow: Direct Implementation
## When to Use
| Criteria | Match |
|----------|-------|
| Intention | simple_change, bug_fix (known cause) |
| Complexity | single file, known solution, low risk |
| Power Mode | Eco or Balanced |
| Clarity | requirements are precise and unambiguous |
## Required Skills
- file-operations
- code-execution
- code-quality-testing
- communication-tone
- safety-profiles (default)
## Flow
### Step 0: Thinking Stage (Pre-Action)
Before any action, apply the Universal Pre-Action Protocol from AGENT.md §B:
- [ ] Do I understand the user's actual intent? Re-evaluate on every message *(Factory)*
- [ ] Have I read the relevant files? NEVER edit code you haven't read
- [ ] Have I verified my assumptions? Search/grep/query to confirm before acting
- [ ] Do I have enough context, or am I guessing?
- [ ] If unsure: search first, ask user second
### Step 1: Understand
- Read the request once
- If unclear: ask ONE clarifying question, then proceed
- If clear: acknowledge and move to execute
### Step 2: Locate
- Search/grep for relevant files
- Read the file(s) you need to modify
- Do NOT explore unrelated code
### Step 3: Edit
- Make the minimal change to satisfy the request
- Use SEARCH/REPLACE for targeted edits
- Combine all changes into minimal tool calls
- Verify: all imports, deps, and syntax correct
### Step 4: Verify
- Run linter on changed files
- Run relevant tests
- Fix any introduced errors (max 3 cycles)
### Step 4b: Sanity Check (Pre-Presentation)
Before presenting, run the Sanity Check from AGENT.md §B Stage 2:
- [ ] Is this the MINIMAL change needed? No scope creep?
- [ ] Does the code have all necessary imports and dependencies?
- [ ] Will it compile/build immediately? *(Windsurf)*
- [ ] Have I checked for linter errors? *(Cursor: max 3 fix cycles)*
- [ ] Did I verify the change fixes the actual problem?
If any check fails: STOP, fix, then proceed.
### Step 5: Present
- Brief summary: what changed, why, and (if applicable) how to verify
- Don't elaborate beyond what was asked
## Token Budget
- **Typical**: 2K-10K tokens
- **Max recommended**: 20K tokens
- **Warning threshold**: 50K tokens → you're probably in the wrong workflow
## GitNexus Integration
- Optional: `query({search_query})` for quick symbol lookup if unsure of location
- Skip for known locations
## Hallucination Watchpoints
- Overcomplicating a simple change (adding features not requested)
- Editing files without reading them first
- Making changes beyond the scope of the request
- Refactoring unrelated code "while you're in there"
## Example
```
User: "Change the port from 3000 to 8080 in server.js"
1. grep for "3000" in server.js → found at line 12
2. read server.js → understand context
3. edit: port 3000 → port 8080
4. run linter → clean
5. "Changed port from 3000 to 8080 in server.js:12"
```