- Home
- Skills
- Product Management
- Repo Audit Planner: Evidence-Checked Fix Plans
More screenshots
Works with the AI Agents you already use
Repo Audit Planner: Evidence-Checked Fix Plans
Audit any repo and get ranked, evidence-checked work orders that a cheaper agent can execute without asking questions.
Free
Repo Audit Planner: Evidence-Checked Fix Plans
Example session with this skill installed
Audit this repo and write plans for the top problems.
- Read your context and instructions
- Compiled the repo audit planner
1. Scope. Covered: the whole repo, 5 files, 54 lines of Python (repo_facts.py: not truncated, no binary or oversized files). Not covered: nothing was skipped. The repo has no dependencies, no CI config and no lockfile, so there was nothing to audit there.
2. Ranked plans.
| Rank | Plan | Title | Severity | Effort | Priority | Status |
|---|---|---|---|---|---|---|
| 1 | 001 | Stop storing money as float in the demo shop | high | S | 4 | todo |
Recommended order: just 001. It computes totals and tax with binary floats, so 0.1 + 0.2 totals 0.30000000000000004 and a tax of 10.825 rounds to 10.82 instead of 10.83. That is wrong money on every order.
3. Rejected candidates.
shop/orders.py:8has a note about discount codes. It is a feature someone wants, not a defect, so it is not a finding.- One test file for four source files (ratio 0.25). The three tests pass and cover pricing and checkout. I found no untested behaviour that is wrong, so "add more tests" would be generic advice, not a finding.
- No CI and no lockfile. The project is a small fixture with no dependencies, so there is nothing to pin or gate.
4. Baseline. From the repo root:
cd examples/demo-shop && python -m pytest -q
Printed 3 passed in 0.01s. The plan checker printed:
ok 001-use-decimal-for-money.md
1 plan(s) checked: 0 error(s), 0 warning(s)
5. Nothing else changed. git status --short at the start printed nothing. At the end it printed:
?? plans/
6. How to run the plan.
Implement
plans/001-use-decimal-for-money.md. Follow it exactly. Run each Verify line and compare with the expected output. Halt and report at any halt condition. Do not touch files outside the plan's scope.
I did not implement anything. To change code, hand that instruction to an executor.
What the checker does with an invented finding. A bad draft claims a SQL injection at shop/orders.py:14, but orders.py has 10 lines. From the repo root:
python3 check_plans.py plans --repo .
ok 001-use-decimal-for-money.md
FAIL 002-fix-sql-injection-in-orders.md
warning plan [section-missing] missing section '## Halt conditions'
warning plan [tests] Test commands should state 'Expected:' with the exit code and success line
error line 22 [evidence-unresolved] examples/demo-shop/shop/orders.py:14: the file has 10 lines
error line 32 [vague] vague wording 'as needed': name the exact file, value or condition
error line 40 [step-no-verify] step has no 'Verify:' line with a command and its expected output
error line 42 [acceptance] Acceptance criteria needs '- [ ]' checklist items
2 plan(s) checked: 4 error(s), 2 warning(s)
Exit code 1. The draft never reaches an executor.
Connects securely to your tools. The creator never sees your data.
About this skill
Use your best model where judgement counts, and a cheap one for the typing.
This skill makes your agent audit a codebase and write a folder of implementation plans. Each plan is a work order: the goal, the evidence, the exact steps with a verify command after each one, acceptance criteria, test commands, halt conditions and a rollback. A smaller model, or a new team member, can carry it out without asking anything.
How it works. Your agent gathers facts with a bundled script (languages, tests, debt markers, largest files, manifests, CI, secret-like locations, churn), hunts for problems, tries to disprove each one, ranks the survivors by severity against effort, and writes one plan per issue. A second script then validates every plan: it opens the files the plan cites and compares the quoted code with the real line. A third script ranks the plans.
What makes it different
- No invented findings: every finding cites
file:lineand quotes the code. If the code is not there, the plan fails. - No vague plans: missing sections, unfilled placeholders and hedges such as "as needed" are errors.
- Ranked by value, with dependencies respected.
- It never touches your code. The skill declines to implement. It writes only under the plans folder and shows
git statusto prove it. - Safe on any repo: the scripts only read, make no network calls and never print a secret value.
Where it stops. The scripts check form and evidence. They cannot tell whether a fix is the right fix, so plan quality follows the model you run. Secret detection is pattern-based and finds common shapes, not every secret.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Product Management.
- The checker verifies that a plan is complete, concrete and backed by real code. It cannot judge whether the plan is the right fix.
- Reads up to 20,000 files of up to 1 MB each and stops after 20 seconds. It says when it truncated.
- Secret detection is pattern-based.
- Needs Python 3.10 or newer to run the scripts. Without it the skill still audits, but cannot validate.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed today
- Free with an Agensi account
- Software
Python 3.10+
- Optional
Git
git is optional. No packages, no network access.
Permissions
File Scopes
Runs three Python scripts from temporary files. No network access, no environment variables, nothing installed.