fix-ci

Name: fix-ci
Author: minghinmatthewlam

34stars4forks

How to install fix-ci?

minghinmatthewlam/agent-guards/fix-ci

$npx -y skills add minghinmatthewlam/agent-guards --skill fix-ci

Installs into the current project.

Summary

Check GitHub Actions status on the current PR, wait for checks to finish, analyze failures and review comments, fix issues, and repeat until CI passes. Use when a user asks to fix failing CI, check PR status, or get a PR merge-ready.

Files · 1

View on GitHub

SKILL.md

1	Usage:
2
3	- `/fix-ci` - Fix CI failures on current branch's PR
4	- `/fix-ci <pr-number>` - Fix specific PR
5	- `/fix-ci --check-only` - Only report status, don't fix
6
7	Instructions:
8
9	Automated CI fix loop. Wait for checks, inspect failures, fix, push, repeat until green.
10
11	Step 1: Parse Input
12
13	- Input: $ARGUMENTS
14	- If a PR number is provided, use that
15	- If `--check-only` flag is present, only report status without fixing
16	- If no arguments, detect PR for current branch via `gh pr view --json number,url`
17
18	Step 2: Readiness Gate
19
20	Before inspecting, ensure CI checks have finished running. Poll every 2 minutes (max 10 iterations / 20 minutes).
21
22	```bash
23	# Count checks still running
24	gh pr checks <pr-number> --json state --jq '[.[] \| select(.state == "PENDING" or .state == "IN_PROGRESS" or .state == "QUEUED")] \| length'
25	```
26
27	- If 0 → proceed to Step 3
28	- If > 0 → wait 2 minutes, retry
29	- After 10 retries → exit with timeout, report which checks never completed
30
31	On subsequent iterations (after pushing a fix): also wait for CI to re-run on the new commit before inspecting again.
32
33	Step 3: Inspect Failures
34
35	Use the bundled inspection script for robust log extraction:
36
37	```bash
38	python "<path-to-skill>/scripts/inspect_pr_checks.py" --repo "." --pr "<pr-number>" --json
39	```
40
41	The script handles:
42	- `gh` CLI field drift (auto-detects available fields, retries with fallbacks)
43	- Smart log extraction (finds failure markers, extracts context window)
44	- Job-level log fallback when run logs aren't available yet
45	- External check detection (non-GitHub Actions checks reported as URL-only)
46
47	If the script is unavailable, fall back to manual inspection:
48
49	```bash
50	gh pr checks <pr-number> --json name,state,conclusion,detailsUrl
51	gh run view <run-id> --log-failed
52	```
53
54	External checks (Buildkite, CircleCI, etc.): report the URL only, mark as out-of-scope. Don't attempt to fetch logs from non-GitHub Actions providers.
55
56	Step 4: Analyze and Fix
57
58	For each failure, read the log snippet and determine the fix. Let the repo context guide you — read project config files, package.json, Makefile, etc. to find the right commands. Don't assume specific tooling.
59
60	General patterns:
61	- Lint errors: Read failing files, fix issues, run the project's lint command to verify
62	- Type errors: Fix type mismatches, run the project's type-check command
63	- Test failures: Read test + implementation files, fix the root cause, run the specific test
64	- Generated file drift: Run the project's generation commands, stage the output
65
66	If `--check-only`, skip this step — just report findings and exit.
67
68	Step 5: Commit and Push
69
70	```bash
71	git add <specific-files>
72	git commit -m "fix: resolve CI failures
73
74	- Fixed [describe what was fixed]
75	- Resolves failures in [job names]"
76	git push
77	```
78
79	Step 6: Loop Until Success
80
81	After pushing, go back to Step 2 (readiness gate). Wait for CI to re-run, then re-inspect.
82
83	- Max 3 iterations
84	- If still failing after 3 iterations → report remaining failures, ask for help
85
86	Step 7: Check PR Review Comments
87
88	After CI passes (or in parallel with fixing), fetch review feedback from all sources:
89
90	```bash
91	# Code-level inline comments
92	gh api "repos/{owner}/{repo}/pulls/<pr-number>/comments" > /tmp/pr_review_comments.json
93
94	# PR-level review bodies
95	gh api "repos/{owner}/{repo}/pulls/<pr-number>/reviews" > /tmp/pr_reviews.json
96
97	# General PR conversation
98	gh pr view <pr-number> --comments
99	```
100
101	Triage each actionable comment into tiers:
102	- Fix — real bug, security issue, logic error, correctness problem
103	- Consider — reasonable improvement, not critical
104	- Skip — nitpick, stylistic preference, bot noise
105
106	Human reviewer comments weighted higher than bot comments. Present the triage to the user and ask which to address. After fixing approved comments, commit, push, and loop back to Step 2.
107
108	Step 8: Report Results
109
110	```
111	## CI Fix Results: PR #<number>
112
113	Status: PASSED \| FAILED \| IN_PROGRESS
114	Branch: <branch> \| URL: <pr-url>
115
116	### Actions Taken
117	1. [Fix description]
118	2. [Fix description]
119
120	### Review Comments
121	- [Summary of addressed/skipped comments]
122
123	### Next Steps
124	[If still failing: what needs manual intervention]
125	```
126
127	Error Handling:
128	- `gh` not authenticated → inform user to run `gh auth login`
129	- No PR exists → inform user to create PR first
130	- Max iterations reached → report remaining failures
131	- External checks failing → report URL, mark out-of-scope
132
133	## Bundled Resources
134
135	### scripts/inspect_pr_checks.py
136
137	Robust CI check inspector. Handles gh CLI field drift, smart log context extraction, job-level log fallbacks, and external check detection.
138
139	```bash
140	python "<path-to-s

People who liked this also viewed

6 picks

Type

fix-ci

34stars4forks

View on GitHub

How to install fix-ci?

minghinmatthewlam/agent-guards/fix-ci

$npx -y skills add minghinmatthewlam/agent-guards --skill fix-ci

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	Usage:
2
3	- `/fix-ci` - Fix CI failures on current branch's PR
4	- `/fix-ci <pr-number>` - Fix specific PR
5	- `/fix-ci --check-only` - Only report status, don't fix
6
7	Instructions:
8
9	Automated CI fix loop. Wait for checks, inspect failures, fix, push, repeat until green.
10
11	Step 1: Parse Input
12
13	- Input: $ARGUMENTS
14	- If a PR number is provided, use that
15	- If `--check-only` flag is present, only report status without fixing
16	- If no arguments, detect PR for current branch via `gh pr view --json number,url`
17
18	Step 2: Readiness Gate
19
20	Before inspecting, ensure CI checks have finished running. Poll every 2 minutes (max 10 iterations / 20 minutes).
21
22	```bash
23	# Count checks still running
24	gh pr checks <pr-number> --json state --jq '[.[] \| select(.state == "PENDING" or .state == "IN_PROGRESS" or .state == "QUEUED")] \| length'
25	```
26
27	- If 0 → proceed to Step 3
28	- If > 0 → wait 2 minutes, retry
29	- After 10 retries → exit with timeout, report which checks never completed
30
31	On subsequent iterations (after pushing a fix): also wait for CI to re-run on the new commit before inspecting again.
32
33	Step 3: Inspect Failures
34
35	Use the bundled inspection script for robust log extraction:
36
37	```bash
38	python "<path-to-skill>/scripts/inspect_pr_checks.py" --repo "." --pr "<pr-number>" --json
39	```
40
41	The script handles:
42	- `gh` CLI field drift (auto-detects available fields, retries with fallbacks)
43	- Smart log extraction (finds failure markers, extracts context window)
44	- Job-level log fallback when run logs aren't available yet
45	- External check detection (non-GitHub Actions checks reported as URL-only)
46
47	If the script is unavailable, fall back to manual inspection:
48
49	```bash
50	gh pr checks <pr-number> --json name,state,conclusion,detailsUrl
51	gh run view <run-id> --log-failed
52	```
53
54	External checks (Buildkite, CircleCI, etc.): report the URL only, mark as out-of-scope. Don't attempt to fetch logs from non-GitHub Actions providers.
55
56	Step 4: Analyze and Fix
57
58	For each failure, read the log snippet and determine the fix. Let the repo context guide you — read project config files, package.json, Makefile, etc. to find the right commands. Don't assume specific tooling.
59
60	General patterns:
61	- Lint errors: Read failing files, fix issues, run the project's lint command to verify
62	- Type errors: Fix type mismatches, run the project's type-check command
63	- Test failures: Read test + implementation files, fix the root cause, run the specific test
64	- Generated file drift: Run the project's generation commands, stage the output
65
66	If `--check-only`, skip this step — just report findings and exit.
67
68	Step 5: Commit and Push
69
70	```bash
71	git add <specific-files>
72	git commit -m "fix: resolve CI failures
73
74	- Fixed [describe what was fixed]
75	- Resolves failures in [job names]"
76	git push
77	```
78
79	Step 6: Loop Until Success
80
81	After pushing, go back to Step 2 (readiness gate). Wait for CI to re-run, then re-inspect.
82
83	- Max 3 iterations
84	- If still failing after 3 iterations → report remaining failures, ask for help
85
86	Step 7: Check PR Review Comments
87
88	After CI passes (or in parallel with fixing), fetch review feedback from all sources:
89
90	```bash
91	# Code-level inline comments
92	gh api "repos/{owner}/{repo}/pulls/<pr-number>/comments" > /tmp/pr_review_comments.json
93
94	# PR-level review bodies
95	gh api "repos/{owner}/{repo}/pulls/<pr-number>/reviews" > /tmp/pr_reviews.json
96
97	# General PR conversation
98	gh pr view <pr-number> --comments
99	```
100
101	Triage each actionable comment into tiers:
102	- Fix — real bug, security issue, logic error, correctness problem
103	- Consider — reasonable improvement, not critical
104	- Skip — nitpick, stylistic preference, bot noise
105
106	Human reviewer comments weighted higher than bot comments. Present the triage to the user and ask which to address. After fixing approved comments, commit, push, and loop back to Step 2.
107
108	Step 8: Report Results
109
110	```
111	## CI Fix Results: PR #<number>
112
113	Status: PASSED \| FAILED \| IN_PROGRESS
114	Branch: <branch> \| URL: <pr-url>
115
116	### Actions Taken
117	1. [Fix description]
118	2. [Fix description]
119
120	### Review Comments
121	- [Summary of addressed/skipped comments]
122
123	### Next Steps
124	[If still failing: what needs manual intervention]
125	```
126
127	Error Handling:
128	- `gh` not authenticated → inform user to run `gh auth login`
129	- No PR exists → inform user to create PR first
130	- Max iterations reached → report remaining failures
131	- External checks failing → report URL, mark out-of-scope
132
133	## Bundled Resources
134
135	### scripts/inspect_pr_checks.py
136
137	Robust CI check inspector. Handles gh CLI field drift, smart log context extraction, job-level log fallbacks, and external check detection.
138
139	```bash
140	python "<path-to-s

People who liked this also viewed

6 picks

Type