Skip to content

Commit 56fddec

Browse files
bartek-aigorabartek-aigora
committed
chore: add develop Claude Code skill
1 parent 37f0300 commit 56fddec

File tree

1 file changed

+288
-0
lines changed

1 file changed

+288
-0
lines changed

‎.claude/skills/develop/SKILL.md‎

Lines changed: 288 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,288 @@
1+
---
2+
name: develop
3+
description: Autonomous development agent. Takes a plan, implements it, self-reviews using Command Center's review APIs, iterates until merge-ready, and sends Slack notification.
4+
argument-hint: <plan text or path to plan file>
5+
---
6+
7+
# Autonomous Development Skill
8+
9+
You are an autonomous development agent. You accept a plan, implement it fully, then self-review using Command Center's review APIs. You iterate on feedback until all gates pass, then notify via Slack.
10+
11+
**IMPORTANT: You MUST keep going until all review gates pass. There is NO iteration limit — you are done when YOU decide the code is ready for human review (all 3 gates pass). Do NOT stop after receiving review feedback — fix the issues and re-review.**
12+
13+
## Phase 1: Setup & Context Detection
14+
15+
1. Resolve configuration. Check these sources in order for each value:
16+
17+
**APP_URL** (Command Center endpoint):
18+
- Default: `https://command-center.aigora.ai`
19+
- Override: `CC_APP_URL` env var, or `APP_URL` in `.env.local`
20+
21+
**API_SECRET** (service-to-service auth):
22+
- Shell env var: `CC_API_SECRET`
23+
- Fallback: `INTERNAL_API_SECRET` in `.env.local`
24+
- If neither exists, tell the user to set `CC_API_SECRET` in their shell profile and stop.
25+
26+
```bash
27+
echo "${CC_API_SECRET:-$(grep INTERNAL_API_SECRET .env.local 2>/dev/null | cut -d= -f2-)}"
28+
```
29+
30+
2. Detect repo from git:
31+
```bash
32+
git remote get-url origin 2>/dev/null | sed 's|.*github.com[:/]||;s|\.git$||'
33+
```
34+
35+
3. Read the plan from `$ARGUMENTS`:
36+
- If `$ARGUMENTS` looks like a file path (ends in `.md`, contains `/`, or starts with `.`), read the file contents and use that as the plan.
37+
- Otherwise, use `$ARGUMENTS` directly as the plan text.
38+
- If `$ARGUMENTS` is empty, tell the user to provide a plan and stop.
39+
40+
4. Summarize the plan into a concise mission goal (2-3 sentences) for use in review API calls.
41+
42+
## Phase 2: Branch & PR Setup
43+
44+
1. Create a feature branch from the current branch:
45+
```bash
46+
git checkout -b feat/<descriptive-name-from-plan>
47+
```
48+
49+
2. Make an initial commit (can be empty or a small scaffold) and push:
50+
```bash
51+
git push -u origin HEAD
52+
```
53+
54+
3. Open a **draft PR** early so PR-dependent review checks can work:
55+
```bash
56+
gh pr create --draft --title "<concise title from plan>" --body "<plan summary>"
57+
```
58+
59+
4. Capture the PR number:
60+
```bash
61+
gh pr view --json number -q .number
62+
```
63+
64+
## Phase 3: Development
65+
66+
Implement the plan using your normal coding tools (Read, Edit, Write, Bash).
67+
68+
- Work through each item in the plan systematically.
69+
- Commit and push after each meaningful chunk of work.
70+
- Use descriptive commit messages that reference what plan item is being addressed.
71+
- Run any available tests locally before pushing.
72+
73+
After all plan items are implemented, push all remaining changes:
74+
```bash
75+
git add <specific-files>
76+
git commit -m "feat: <description of changes>"
77+
git push
78+
```
79+
80+
## Phase 4: Review Loop
81+
82+
Before calling any review APIs, do a **self-assessment**: review the plan and confirm all items are implemented. If something is missing, go back to Phase 3.
83+
84+
Then run the three review gates in order. All three must pass to exit the loop. **There is no iteration limit** — keep fixing and re-reviewing until all gates pass.
85+
86+
### Gate 1: Code Quality
87+
88+
```bash
89+
curl -s -X POST "${APP_URL}/api/review/code-quality" \
90+
-H "Content-Type: application/json" \
91+
-H "x-internal-api-secret: ${API_SECRET}" \
92+
-d '{"repo":"REPO","branch":"BRANCH","missionGoal":"MISSION_GOAL"}'
93+
```
94+
95+
Parse with python3:
96+
```bash
97+
python3 -c "
98+
import sys, json
99+
data = json.load(sys.stdin)
100+
if not data.get('success'):
101+
print('ERROR:', data.get('error', 'Unknown error'))
102+
sys.exit(1)
103+
r = data['result']
104+
print('PASSED:', r.get('passed', False))
105+
print('ASSESSMENT:', r.get('overallAssessment', 'unknown'))
106+
print('---FINDINGS---')
107+
findings = r.get('findings', '')
108+
if isinstance(findings, str):
109+
for f in findings.split('\\n\\n'):
110+
if f.strip(): print(f); print()
111+
print('---AGENT_INSTRUCTIONS---')
112+
print(r.get('agentInstructions', 'None'))
113+
"
114+
```
115+
116+
**Pass criteria:** `passed: true`
117+
**On failure:** Fix issues using the `findings` and `agentInstructions` from the response. Commit, push, and re-run this gate.
118+
119+
### Gate 2: PR Review
120+
121+
```bash
122+
curl -s -X POST "${APP_URL}/api/review/pr-review" \
123+
-H "Content-Type: application/json" \
124+
-H "x-internal-api-secret: ${API_SECRET}" \
125+
-d '{"repo":"REPO","prNumber":PR_NUM,"missionGoal":"MISSION_GOAL"}'
126+
```
127+
128+
Parse with python3:
129+
```bash
130+
python3 -c "
131+
import sys, json
132+
data = json.load(sys.stdin)
133+
if not data.get('success'):
134+
print('ERROR:', data.get('error', 'Unknown error'))
135+
sys.exit(1)
136+
r = data['result']
137+
print('OPEN_ITEMS:', r.get('openItemCount', 0))
138+
print('AI_FINDINGS:', r.get('aiFindingCount', 0))
139+
print('---ACTION_ITEMS---')
140+
for item in r.get('actionItems', []):
141+
print('-', item)
142+
print('---FEEDBACK---')
143+
print(r.get('actionableFeedback', 'None'))
144+
"
145+
```
146+
147+
**Pass criteria:** `openItemCount: 0` (no unresolved critical/high items)
148+
**On failure:** Fix issues using the `actionableFeedback` and `actionItems`. Commit, push, and re-run this gate.
149+
150+
### Gate 3: Merge Readiness
151+
152+
```bash
153+
curl -s -X POST "${APP_URL}/api/review/merge-readiness" \
154+
-H "Content-Type: application/json" \
155+
-H "x-internal-api-secret: ${API_SECRET}" \
156+
-d '{"repo":"REPO","branch":"BRANCH","prNumber":PR_NUM}'
157+
```
158+
159+
Parse with python3:
160+
```bash
161+
python3 -c "
162+
import sys, json
163+
data = json.load(sys.stdin)
164+
if not data.get('success'):
165+
print('ERROR:', data.get('error', 'Unknown error'))
166+
sys.exit(1)
167+
r = data['result']
168+
print('IS_READY:', r.get('isReady', False))
169+
print('AI_VERDICT_READY:', r.get('aiVerdictReady', False))
170+
print('---BLOCKING---')
171+
for issue in r.get('blockingIssues', []):
172+
print('-', issue)
173+
print('---WARNINGS---')
174+
for w in r.get('warnings', []):
175+
print('-', w)
176+
"
177+
```
178+
179+
**Pass criteria:** `aiVerdictReady: true`
180+
**On failure:** Fix issues using the `blockingIssues`. Commit, push, and re-run the failing gate.
181+
182+
### Handling Findings — Including Pre-Existing Issues
183+
184+
The review APIs analyze the full branch, not just your diff. This means they may surface **pre-existing issues** in code you didn't write. This is intentional — the codebase should improve over time with every PR.
185+
186+
**Triage each finding into one of three categories:**
187+
188+
1. **Genuine issue (your code or pre-existing)** — Fix it. Real bugs, security issues, missing error handling, and code quality problems should be fixed regardless of whether you introduced them. Leave the codebase better than you found it.
189+
190+
2. **Intentional design choice / acceptable trade-off** — The code is correct but the AI disagrees with the pattern. For example: a deliberate fallback to env vars, a purposely loose type, or a known limitation documented in comments. These are not actionable.
191+
192+
3. **AI hallucination / stale data** — The finding references code that doesn't exist, misreads the logic, or flags a review comment that was already resolved. These are false positives.
193+
194+
**For categories 2 and 3:**
195+
- Track them across runs. If the same non-actionable findings persist for **3 consecutive runs** after you've inspected the code and confirmed they don't warrant changes:
196+
- **Stop the review loop.**
197+
- Tell the user exactly which findings are persisting, your assessment of each (why it's intentional or a false positive), and ask for guidance on whether to proceed to finalization or make additional changes.
198+
- **Never make unnecessary code changes** just to appease a finding you believe is wrong. Unnecessary changes introduce risk and noise in the PR.
199+
200+
### After fixing any gate
201+
202+
1. Commit and push the fixes:
203+
```bash
204+
git add <specific-files-you-changed>
205+
git commit -m "fix: address review feedback - <brief description>"
206+
git push
207+
```
208+
209+
2. Re-run the **failing gate** (not all gates — only repeat from the gate that failed).
210+
211+
3. Once a gate passes, move to the next gate.
212+
213+
4. If all 3 gates pass, exit the review loop.
214+
215+
5. Keep iterating until all 3 gates pass. You decide when the code is ready — there is no artificial limit. The only exception is persistent false positives (see above), where you should escalate to the user.
216+
217+
## Phase 5: Finalize & Notify
218+
219+
1. Mark the PR as ready for review (remove draft status):
220+
```bash
221+
gh pr ready
222+
```
223+
224+
2. Send a Slack notification:
225+
```bash
226+
PR_URL=$(gh pr view --json url -q .url)
227+
curl -s -X POST "${APP_URL}/api/review/notify" \
228+
-H "Content-Type: application/json" \
229+
-H "x-internal-api-secret: ${API_SECRET}" \
230+
-d "{\"type\":\"ready_for_review\",\"repo\":\"REPO\",\"branch\":\"BRANCH\",\"data\":{\"prUrl\":\"${PR_URL}\",\"missionGoal\":\"MISSION_GOAL\"}}"
231+
```
232+
233+
3. Tell the user:
234+
> All review gates passed. PR is ready for human review.
235+
> **PR:** [PR_URL]
236+
> **Branch:** BRANCH
237+
> **Slack notification sent.**
238+
239+
## Available Review Tools Reference
240+
241+
### 1. Code Quality (`POST /api/review/code-quality`)
242+
AI-powered code quality analysis. Does NOT require a PR — reviews the branch directly.
243+
244+
**Request:** `{"repo": "owner/repo", "branch": "feat/my-feature", "missionGoal": "what the code should accomplish"}`
245+
246+
**Key response fields:**
247+
- `result.passed` (boolean) — whether quality meets the bar
248+
- `result.overallAssessment` — "healthy", "needs-attention", or "critical"
249+
- `result.findings` (string) — detailed findings markdown
250+
- `result.agentInstructions` (string) — specific fix instructions for agents
251+
252+
### 2. PR Review (`POST /api/review/pr-review`)
253+
AI-powered analysis of PR review comments. Checks for unresolved or critical feedback from reviewers.
254+
255+
**Request:** `{"repo": "owner/repo", "prNumber": 42, "missionGoal": "what the code should accomplish"}`
256+
257+
**Key response fields:**
258+
- `result.openItemCount` (number) — count of unresolved critical/high items
259+
- `result.aiFindingCount` (number) — count of AI-generated findings
260+
- `result.actionItems` (string[]) — list of specific actions to take
261+
- `result.actionableFeedback` (string) — formatted feedback markdown
262+
263+
### 3. Merge Readiness (`POST /api/review/merge-readiness`)
264+
Comprehensive merge readiness check. Requires a PR. Analyzes diff, reviews, docs, and build status.
265+
266+
**Request:** `{"repo": "owner/repo", "branch": "feat/my-feature", "prNumber": 42}`
267+
268+
**Key response fields:**
269+
- `result.isReady` (boolean) — overall readiness
270+
- `result.aiVerdictReady` (boolean) — AI's merge verdict
271+
- `result.blockingIssues` (string[]) — issues that must be fixed
272+
- `result.warnings` (string[]) — non-blocking warnings
273+
274+
### 4. Notify (`POST /api/review/notify`)
275+
Send a Slack notification. Use after all checks pass.
276+
277+
**Request:** `{"type": "ready_for_review", "repo": "owner/repo", "branch": "feat/my-feature", "data": {"prUrl": "https://...", "missionGoal": "..."}}`
278+
279+
## Important Notes
280+
281+
- **Do NOT use `jq`** — it may not be available. Use `python3 -c "import sys,json; ..."` for JSON parsing.
282+
- **Draft PR opened early** so PR-dependent checks (pr-review, merge-readiness) work from the start.
283+
- **Self-assess before API calls** to avoid burning expensive AI calls on incomplete work.
284+
- **No iteration limit** — keep going until all gates pass. You decide when the code is ready for human review.
285+
- **Each review call takes 30-90 seconds** — tell the user to expect a wait.
286+
- **Push before re-reviewing** — the review APIs check the latest commit on the branch.
287+
- **Be thorough with fixes** — superficial fixes will just fail the next review cycle.
288+
- **All auth headers** use `x-internal-api-secret` with the resolved `API_SECRET` value.

0 commit comments

Comments
 (0)