docs: remove internal-only e2e role from the user-facing role matrix

[waynesun09]

What

Removes the e2e row and its permissions note from the Role Permissions Matrix in docs/guides/infrastructure/infrastructure-reference.md (the row was added in #6652).

Why

e2e is an internal role used only to drive fullsend's own end-to-end / behaviour test tooling against disposable test orgs. It is not a role users select for their agents. Listing it in the user-facing matrix alongside coder/review/triage/etc. implies it is a selectable option, which is exactly the role-selection confusion #6563 is about. The matrix should show only roles intended for user/agent configuration.

This is docs-only — no behaviour change. The role continues to exist and function for internal test tooling.

Relates to #6563 (BYOA role selection).

🤖 Generated with Claude Code

[qodo-code-review]

Code Review by Qodo

🐞 Bugs (0) 📘 Rule violations (0) 📎 Requirement gaps (0)

Great, no issues found!

Qodo reviewed your code and found no material issues that require review

Tip of the day

💡 Did you know, you can group findings by type and pick your Finding display, from Minimal to Full

More tips ↗ | Customize Qodo ↗ | Qodo docs ↗

[qodo-code-review]

PR Summary by Qodo

Remove internal e2e role from the user-facing permissions matrix

📝 Documentation 🕐 Less than 5 minutes

AI Description

• Removes the internal-only e2e role from the user-facing permissions matrix.
• Removes the related note describing permissions irrelevant to configurable agent roles.
• Prevents readers from mistaking e2e for a selectable agent role.

High-Level Assessment

Removing the internal role from the user-facing matrix is the clearest approach because it preserves internal test behavior while eliminating role-selection ambiguity. Adding an internal-only annotation was considered but would retain irrelevant implementation detail in user documentation.

Files changed (1) +0 / -3

Documentation (1) +0 / -3

infrastructure-reference.mdRemove internal e2e role from permissions matrix +0/-3

Remove internal e2e role from permissions matrix

• Removes the 'e2e' role row and its additional permissions note from the Role Permissions Matrix. The matrix now lists only roles intended for user and agent configuration.

docs/guides/infrastructure/infrastructure-reference.md

[github-actions]

Site preview

Preview: https://012f6da0-site.fullsend-ai.workers.dev

Commit: 025ad23a861b7bada3c26b029ce8f7bd6ca56933

[fullsend-ai-review]

🤖 Finished Review · ✅ Success · Started 10:23 PM UTC · Completed 10:35 PM UTC

Commit: 025ad23 · View workflow run →

Runtime: claude · Model: opus → claude-opus-4-6 · Effort: high · Cost: $3.12

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[fullsend-ai-review]

Risk Assessment: low (1/5)

Details

Docs-only change: removes the internal e2e role row (3 lines) from the user-facing role permissions matrix in infrastructure-reference.md. Single file, no code, no protected paths, no security sensitivity. Aligns with linked issue #6563 about confusing internal roles. Trivially reversible.

[fullsend-ai-review]

Review

Findings

Medium

• [stale-doc-reference] docs/guides/user/bring-your-own-agent.md:338 — The BYO agent guide still lists e2e as a valid value in the roles field: "Valid values: fullsend, triage, coder, review, fix, retro, prioritize, e2e". This is a user-facing doc and contradicts the PR's goal of removing e2e from user-facing documentation.
  Remediation: Remove e2e from the valid-values list, or add a note that it is internal-only.

Low

• [stale-doc-reference] docs/cli/mint.md:108 — The --roles flag example in the CLI reference still includes e2e in its comma-separated list: --roles=fullsend,triage,coder,review,retro,prioritize,e2e. While this is an operator-facing doc, consistency with the PR's intent would suggest updating it too.
  Remediation: Remove ,e2e from the example value, or keep it intentionally if this doc targets internal operators.

---

Next steps:

• /fs-fix — agent addresses review findings automatically
• /fs-fix <your instruction> — agent fixes with your specific guidance
• Push commits directly — review re-runs automatically on push
• /fs-fix-stop — disable automatic fix runs for this PR

[fullsend-ai-review]

See the review comment for full details.

[fullsend-ai-retro]

🤖 Finished Retro · ✅ Success · Started 9:37 AM UTC · Completed 9:45 AM UTC

Commit: 025ad23 · View workflow run →

Runtime: claude · Model: opus → claude-opus-4-6 · Effort: high · Cost: $2.17

[fullsend-ai-retro]

Retro: PR #6771 — docs: remove internal-only e2e role from the user-facing role matrix

Timeline

1. 2026-08-24: Retro agent filed issue #6563 about opaque 403 errors for unregistered BYO agent roles.
2. 2026-08-29 22:21 UTC: Wayne Sun opened PR #6771, a docs-only change removing 3 lines from infrastructure-reference.md to eliminate the e2e row from the user-facing role permissions matrix. Co-authored with Claude Opus 4.8 via a local Claude Code session.
3. 2026-08-29 22:22–22:35 UTC: Review agent ran (workflow run 33278403494) — Opus, high effort, $3.12. Assessed risk as low (1/5). Requested changes with two findings: (a) bring-your-own-agent.md:338 still lists e2e as a valid role value (medium), (b) mint.md:108 CLI example still includes e2e (low).
4. 2026-08-31 06:35 UTC: Human reviewer rh-hemartin approved without comments.
5. 2026-09-01 09:35 UTC: PR merged without addressing the review agent's findings.

What went well

The review agent's cross-file stale-reference detection was excellent. It correctly identified that removing e2e from one docs file while leaving it in two other user-facing docs creates inconsistency. This is the kind of mechanical consistency check where automated review consistently outperforms human review. The agent's medium-severity finding about bring-your-own-agent.md is particularly strong — that file is explicitly user-facing and listing e2e as a valid role value directly contradicts the PR's stated goal of preventing users from thinking e2e is a selectable role.

Evidence for existing issues

• #6666 (stale Sonnet model version mapping): During this review, the agent attempted to delegate sub-agents to Sonnet to reduce cost but failed — the model was not available. It wasted ~1 minute of wall time probing environment variables before falling back to Opus for all sub-agents. This is consistent with Stale sonnet model version mapping inflates review sub-agent costs ~5x #6666's description of stale model version mappings.
• #5777 (review model routing by diff complexity): $3.12 on Opus at high effort for a 3-line, single-file docs-only change is disproportionate. This PR is a clear candidate for Sonnet-tier routing — the agent's actual findings (stale cross-file references) required thoroughness but not Opus-level reasoning.
• #4936 / #3377 (autonomy readiness for stale-reference detection): This PR provides additional evidence that the review agent's cross-codebase stale-reference detection exceeds human review quality. The agent found two valid stale references that the human reviewer did not address. This is consistent with the pattern tracked in Track autonomy readiness: review agent cross-codebase stale-reference detection exceeded human review on build-definitions #3688 #4936 (build-definitions) and Track autonomy readiness: review agent outperformed human review on mechanical consistency in documentation repo #3377 (documentation repo).
• #2449 (consolidate agent role lists): The root cause of the stale e2e references is that role names are defined in multiple places with no single source of truth. The tactical fix proposed below cleans up the docs, but Consolidate agent role lists and permission definitions #2449 addresses the structural problem.

Proposals filed

• Remove stale e2e role references from user-facing docs (bring-your-own-agent.md, mint.md) (in fullsend-ai/fullsend)