# Tech Lead Scorecard answer key

URL: https://developertoolkit.ai/en/lead-scorecard-guide/
Description: Each of the Tech Lead Scorecard's 25 questions mapped to the page that moves it, the evidence a 3-point answer needs, and a four-week plan for one team.

The Tech Lead Scorecard answer key maps each of the 25 questions on the [Tech Lead Scorecard](/en/lead-scorecard/) to the page that moves that answer and names the evidence a 3-point answer needs. It is for a tech lead or engineering manager choosing which three gaps to close first and how to prove they closed.

Your result says **Level 2 · Paired**, 31 points out of 75, with the headline "Team Enabler". Two developers run parallel agents and ship clean pull requests; the other six paste into chat and hand you diffs to read. The review queue is the bottleneck, and 25 per-question pages is too much reading before Monday. This key picks the first three questions and tells you what counts as done.

**How the Tech Lead Scorecard scores**

25 single-choice questions, 0 to 3 points each, 75 points in total, in six sections: shared standards (12 points), review and quality gates (12), rollout and adoption (15), team-scale workflow (12), enablement (12), and team ops (12). The results page adds a top-risks list, a 30/60/90-day plan once you add your email, and an **Improve these answers** card that links the page for every answer below 3. Those links are the **Start with** column below.

## What does each Level 1–4 band mean for a team?

The bands borrow the names of the [autonomy ladder](/en/ladder/), but they measure something else: the practices around one team's agent use, as you report them. The ladder measures how each loop ships merged change. [One map](/en/ladder/one-map/) shows how to place a team by what actually reaches production, so use it to check your self-reported band. Answer for the team's median developer, not its strongest one.

| Band | Points | Headline on the results page | First move |
| --- | --- | --- | --- |
| **Level 1 · Assisted** | 0–18 | Lone Wolf: it works for you, now make it work for the team | Section 1: commit one shared, versioned rules file to the most active repository. |
| **Level 2 · Paired** | 19–37 | Team Enabler: pockets of consistency, connect them | Section 2: the same automated gates on every pull request, whoever or whatever wrote it. |
| **Level 3 · Review manager** | 38–56 | Force Multiplier: the team follows one way of working | Sections 4 and 6: parallel agents with a concurrency cap, and cost and secrets under an owner. |
| **Level 4 · Spec manager** | 57–75 | Org Amplifier: a self-sustaining team system | Teach the system to another team. [Moving the team off reading every diff](/en/teams/trust-transfer/) is the work that remains. |

## How do you pick the first three questions to fix?

1. Divide each section's points by its maximum. A 6 out of 15 in rollout (40%) is a bigger gap than 6 out of 12 in enablement (50%).
2. Take the two sections with the lowest share. In each, pick the question where your answer is furthest from 3 points.
3. If review and quality gates (questions 5–8) sit below half, make one of the three picks a question from that section, even when another section scores lower. More agent output through weak gates only lengthens [the review queue](/en/teams/review-queue/). For the same reason, raise team-scale workflow (questions 14–17) only after the gates hold.
4. For each pick, write the target answer and the artifact that will prove it into the plan file below, with one owner per question.
5. Re-take the scorecard after four weeks. Compare section shares, not the band: one question moving from 1 to 3 points rarely changes the band, but it shows in its section.

Commit the plan to the team's main repository, next to the shared agent rules, so it is reviewed like any other change:

```yaml
# .agents/scorecard-plan.yaml — one entry per question the team is working on
- question: 7          # What automated gates run on AI-authored code?
  current_points: 1    # "Lint only"
  target_points: 3     # "Layered: AI review + CI + coverage gates"
  evidence: "Branch protection requires lint, types, tests, coverage and the review agent; three merged agent PRs show all five green"
  owner: "priya"
  check_on: 2026-10-26
```

## Which page answers each question?

Each table gives the question, the page the results card links (**Start with**), a second page that goes deeper, and the evidence that earns 3 points. Claim a 3 only when you can link that evidence.

### Shared standards and context (questions 1–4, 12 points)

| # | Question | Start with · Go deeper | Evidence for 3 points |
| --- | --- | --- | --- |
| 1 | Does your team share a common agent config? | [Shared agent rules](/en/cto-scorecard-guide/shared-agent-rules/) · [Concise repository context](/en/developer-scorecard-guide/claude-md-context/) | A reviewed `AGENTS.md` or `CLAUDE.md` in every active repository, changed only through pull requests |
| 2 | How is project context available to everyone's agent? | [Documentation as context](/en/workflows/context/documentation-as-context/) · [Team collaboration](/en/teams/adopt/team-collaboration/) | Architecture and convention docs linked from the rules file and updated in the same pull request as the code they describe |
| 3 | Are prompt and workflow patterns shared across the team? | [Shared skills](/en/cto-scorecard-guide/shared-skills/) · [Building custom skills](/en/ecosystem/skills/building-custom-skills/) | A skills or commands directory in the repository, one owner per entry |
| 4 | How fast is a new developer productive with your setup? | [Developer onboarding](/en/cto-scorecard-guide/dev-onboarding-time/) · [The developer track](/en/start/developer/) | One command that sets up a fresh clone, and a written first-week path the last joiner followed |

### Review and quality gates (questions 5–8, 12 points)

| # | Question | Start with · Go deeper | Evidence for 3 points |
| --- | --- | --- | --- |
| 5 | How consistent is AI-generated code quality across the team? | [Code quality gates](/en/workflows/test/code-quality/) · [Fitness functions](/en/workflows/test/fitness-functions/) | The same required checks on every pull request, and architecture rules enforced in CI |
| 6 | Do you have review standards specific to AI-assisted PRs? | [Team PR review automation](/en/cto-scorecard-guide/team-pr-review-automation/) · [Running the review queue](/en/teams/review-queue/) | A written standard for agent pull requests, with its mechanical parts enforced by CI |
| 7 | What automated gates run on AI-authored code? | [Layered pull-request review](/en/developer-scorecard-guide/layered-pr-review/) · [The evidence bundle](/en/workflows/ship/evidence-bundle/) | Branch protection that requires CI, a coverage gate and an AI review before merge |
| 8 | How do you catch plausible-but-wrong code before merge? | [Test: give the session a feedback loop](/en/method/lifecycle/test/) · [Catching plausible-but-wrong code](/en/workflows/test/slop-detection/) | Tests the agent must pass before it stops, plus an adversarial review pass with its own prompt |

### Rollout and adoption (questions 9–13, 15 points)

| # | Question | Start with · Go deeper | Evidence for 3 points |
| --- | --- | --- | --- |
| 9 | What share of the team actively uses agentic tools? | [Team adoption rate](/en/cto-scorecard-guide/team-adoption-rate/) · [Team onboarding and adoption](/en/teams/adopt/team-adoption/) | Daily use by more than 80% of the team, read from a usage report rather than a show of hands |
| 10 | How did adoption happen on your team? | [Adoption roadmap](/en/teams/adopt/adoption-roadmap/) · [Designing a pilot](/en/org/pilot-design/) | A written rollout with a pilot repository, a support owner and an exit criterion |
| 11 | Do you measure the impact of AI? | [AI metrics panel](/en/cto-scorecard-guide/ai-metrics-panel/) · [Metrics frameworks](/en/org/metrics-frameworks/) | A dashboard with a pre-rollout baseline and a goal, reviewed on a fixed date |
| 12 | How do skeptics and senior engineers engage? | [Team onboarding and adoption](/en/teams/adopt/team-adoption/) · [Skeptics and senior engineers](/en/teams/skeptics-and-seniors/) | Senior engineers authored or approved the shared rules and gates |
| 13 | Who owns tooling decisions and budget? | [Tooling policy](/en/cto-scorecard-guide/tooling-policy/) · [Cost governance](/en/org/cost-governance/) | A named owner, a budget line and a review date, written down |

### Team-scale workflow (questions 14–17, 12 points)

| # | Question | Start with · Go deeper | Evidence for 3 points |
| --- | --- | --- | --- |
| 14 | Does the team use parallel agents or worktrees? | [Team parallelism](/en/cto-scorecard-guide/team-parallelism-tier2/) · [Parallel agents in worktrees](/en/developer-scorecard-guide/git-worktrees/) | A documented worktree convention with a concurrency cap per developer |
| 15 | How automated is your PR, review and merge loop? | [A bounded PR review-fix loop](/en/developer-scorecard-guide/auto-pr-workflow/) · [Team PR review automation](/en/cto-scorecard-guide/team-pr-review-automation/) | The agent opens the pull request and fixes review findings for a capped number of rounds; a human merges |
| 16 | Do you use shared MCP servers across the team? | [Internal MCP servers](/en/cto-scorecard-guide/internal-mcp-servers/) · [Essential MCP servers](/en/ecosystem/mcp/essential-servers/) | A checked-in MCP configuration with an owner per server |
| 17 | How do you handle large, cross-cutting changes? | [Million-line codebase strategies](/en/workflows/build/large-codebases/) · [Shaping a backlog for agents](/en/teams/agent-ready-backlog/) | A reviewed plan before execution, and the change split into tickets an agent can finish and verify |

### Enablement and mentoring (questions 18–21, 12 points)

| # | Question | Start with · Go deeper | Evidence for 3 points |
| --- | --- | --- | --- |
| 18 | How do you level up the team's AI skills? | [The developer track](/en/start/developer/) · [Upskilling a team](/en/teams/upskilling/) | A curriculum with practice and review of the output, and a record of who completed it |
| 19 | Do you run internal knowledge-sharing on AI workflows? | [Knowledge sharing](/en/cto-scorecard-guide/knowledge-sharing/) · [Workflow transformation](/en/teams/adopt/workflow-transformation/) | A recurring session on the calendar and the artifacts each one produced |
| 20 | When someone finds a great workflow, how does it spread? | [Shared skills](/en/cto-scorecard-guide/shared-skills/) · [Agent skills](/en/developer-scorecard-guide/agent-skills/) | The workflow lives as a skill in the repository and was taught in a session |
| 21 | How do you keep the team current as tools change? | [AI tooling roadmap](/en/cto-scorecard-guide/ai-tooling-roadmap/) · [Keeping a team current](/en/teams/keeping-current/) | An owner who triages releases, a sandbox to try them, and recorded adopt-or-skip decisions |

### Team ops and governance (questions 22–25, 12 points)

| # | Question | Start with · Go deeper | Evidence for 3 points |
| --- | --- | --- | --- |
| 22 | How do you manage AI cost across the team? | [Cost governance](/en/org/cost-governance/) · [Model routing by evidence](/en/developer-scorecard-guide/model-routing/) | Spend per person or per repository reviewed monthly, plus written model-routing guidance |
| 23 | How are secrets and permissions handled for team agents and MCP? | [MCP security](/en/cto-scorecard-guide/mcp-security/) · [Agent identity and secrets](/en/org/agent-identity-and-secrets/) | Secrets from a secret manager, a scoped token per agent integration, and a written policy |
| 24 | Is there a policy for what AI cannot touch? | [Governance and autonomy](/en/org/governance-autonomy/) · [Hooks as deterministic guardrails](/en/developer-scorecard-guide/agent-hooks/) | Written limits enforced by permission settings or hooks, not only by instructions in a prompt |
| 25 | How do you decide which tools the team standardizes on? | [Tooling policy](/en/cto-scorecard-guide/tooling-policy/) · [Tool comparison](/en/comparison/) | A trial on the team's own tasks, the chosen standard, and a revisit date |

## How do you prove a higher answer without reading every diff?

Every 3-point answer above names an artifact, not an intention: a branch-protection export, a checked-in config, a dashboard with a baseline, merged pull requests that carry an [evidence bundle](/en/workflows/ship/evidence-bundle/). The owner in the plan file produces the artifact, and a second engineer checks it before you re-take the scorecard. [Reading evidence instead of code](/en/ladder/evidence-not-diffs/) explains why that check replaces reading each diff.

An agent can collect most of that evidence from the repository, provided it runs read-only. Do not rely on the prompt alone to keep it read-only; start it in a mode that does not edit files.

**Claude Code**

Start the session in plan mode, which explores without editing files: `claude --permission-mode plan` (checked in `claude --help`, v2.1.283). Without the flag, interactive sessions from v2.1.283 (both release channels on 2026-10-12) usually start in auto mode, which can edit files. Paste a prompt below.

**Codex**

Run the prompt non-interactively in the read-only sandbox: `codex exec --sandbox read-only "PROMPT"` (checked in `codex exec --help`, v0.157.1), where `PROMPT` is one of the prompts below. That is the legacy sandbox flag; the beta permission profiles select the same with `-c default_permissions=":read-only"`. The two systems do not compose, so use one.

**Cursor**

Switch the agent to Plan Mode, which produces a plan before writing any code, and do not approve the build step; paste a prompt below.

The prompt text is the same in all three tools.

**Copy-paste prompt: evidence for standards and gates (questions 1–8)**

```text
Read-only task. Do not edit files, run git commands that write, or call any network API.
I am answering Tech Lead Scorecard questions 1-8 for this repository. List the evidence that exists today:
- Agent rules files (AGENTS.md, CLAUDE.md, .cursor/rules, nested copies): where they are, when each last changed (git log -1), and whether they reference architecture or convention docs
- Shared skills or commands directories, and who is named as owner, if anyone
- A bootstrap script or setup command for a fresh clone
- CI workflow files: each job that runs on pull requests and what makes it fail (lint, types, tests, coverage threshold, AI review)
- PR templates or checks that ask for test evidence or a risk description
Cite each item as path:line. Then, per question, state the answer on the 0-3 scale the evidence supports and what is missing for 3. Write "not found" rather than guessing.
```

**Copy-paste prompt: evidence for team ops (questions 16, 22–24)**

```text
Read-only task. Do not edit files, print secret values, or call any network API.
I am answering Tech Lead Scorecard questions 16, 22, 23 and 24 for this repository. Find:
- MCP configuration files (.mcp.json, .cursor/mcp.json, .codex/config.toml): each server, how it authenticates, and whether a credential appears inline instead of as an environment variable reference
- Agent permission settings (.claude/settings.json, Codex sandbox or permission settings, Cursor rules): what is denied and what is allowed
- Hooks or CI steps that block agent changes to protected paths, production config, migrations or secrets
- Any written guidance on model choice or spend
Cite each item as path:line and name any credential by file and key only. Per question, state the answer on the 0-3 scale the evidence supports and what is missing for 3.
```

Branch protection, required reviewers and usage reports live in your Git host and your tools' admin consoles, not in the repository. Export them and attach them to the same plan entry.

## When does a Tech Lead Scorecard result mislead?

- **You scored intentions.** A review checklist nobody enforces earns the same answer as a required check if you let it. Recovery: re-score every answer you cannot back with a linked artifact one point lower; that total is your real starting point.
- **You answered for your two strongest developers.** Their parallel agents do not make the team Level 3. Recovery: answer for the median developer, and use question 9's usage report to check.
- **Workflow rose before the gates.** More parallel agents and auto-opened pull requests against the same review capacity grow the queue until reviewers approve without looking. Recovery: cap concurrent agent pull requests per developer, fix questions 6–8 first, then raise the cap.
- **Seniors were routed around, not brought in.** Adoption climbs while question 12 stays at 1 and the seniors quietly re-review everything. Recovery: give them ownership of the rules and the gates, as [skeptics and senior engineers](/en/teams/skeptics-and-seniors/) describes.
- **The band became a target.** A jump of a band in four weeks usually means optimistic answers. Recovery: re-check each changed answer against its artifact and report section shares to your manager, not the band.

## Where to go next from your Tech Lead Scorecard result

- [Take the Tech Lead Scorecard](/en/lead-scorecard/) — 25 questions, about 8 minutes: a Level 1–4 band, top risks, and a 30/60/90-day plan.
- [The tech lead track](/en/start/tech-lead/) — The ordered reading path: shared harness, verification design, review capacity.
- [Lead a team](/en/teams/) — Rollout, the shared harness, review and flow, and people: the section this key draws on.
- [CTO Scorecard answer key](/en/cto-scorecard-guide/) — The same decisions at organization scale, for the leader your plan reports to.
