docs: add tool_output_cap.py to STRUCTURE.md, cross-ref in agent-tools [doc-updater]

[james-in-a-box]

Update documentation to reflect changes introduced by bb71f4d (#2877), which added shared/egg_agent/tool_output_cap.py — a new PreToolUse hook that bounds built-in CC tool output.

• STRUCTURE.md: Add tool_output_cap.py to the egg_agent/ module listing. The commit added the file but STRUCTURE.md still showed only tool_interceptor.py as the last entry.
• agent-tools.md: Add a cross-reference from the existing MCP output-size cap section (EGG_TOOL_OUTPUT_CAP_BYTES, Cap MCP @tool output sizes at the tool layer (follow-up to #2804) #2805) to the new built-in tool predictive cap (EGG_TOOL_OUTPUT_CAP / EGG_READ_CAP_BYTES, Bound built-in Claude Code tool output (Read/Edit/Grep) via PreToolUse predictive cap (follow-up to #2805/#2810) #2876). Both mechanisms protect the same 1 MB SDK buffer limit via different layers; the cross-reference helps operators find the complementary knob.

The docs/reference/agent-recovery.md was already updated as part of the original commit and does not need further changes.

Triggered by: #2877

Test Plan

• Automated: doc-only changes, no code modified
• Manual: verify STRUCTURE.md lists tool_output_cap.py and agent-tools.md cross-reference links to agent-recovery.md anchor

Authored-by: egg

[egg-reviewer]

LGTM — doc-only PR, every factual claim verified against the code and target docs. No blocking issues.

What I verified

• tool_output_cap.py exists at shared/egg_agent/tool_output_cap.py, and the STRUCTURE.md one-liner is accurate:
  • "Read/Grep" matches the dispatcher — check_builtin_tool_output_risk only handles Read and Grep (tool_output_cap.py:266-269). This is actually more precise than the module's own docstring, which loosely lists Edit/Write/Bash as built-ins too.
  • "1 MB SDK buffer", EGG_TOOL_OUTPUT_CAP / EGG_READ_CAP_BYTES, and #2876 all match the source (is_output_cap_disabled, _read_cap_bytes, module docstring).
• STRUCTURE.md tree is correct: tool_interceptor.py correctly switched from └── to ├──, the new file is the └── terminal entry, and the listing matches the real egg_agent/ directory contents and alphabetical order.
• agent-tools.md cross-reference resolves: the anchor #predictive-output-cap-pretooluse maps to the real heading ### Predictive Output Cap (PreToolUse) (agent-recovery.md:136) under GitHub's slug rules. The relative path is correct (both files live in docs/reference/).
• Issue references (#2876, #2805, #2804) are consistent with the linked section and the source.
• PR-body claim that agent-recovery.md needs no further changes is correct — that section already exists from the prior commit.

The new agent-tools.md paragraph is placed sensibly right after the complementary EGG_TOOL_OUTPUT_CAP_BYTES (#2805) section, and the "Built-in tools (Read, Grep, etc.) … can't be wrapped" framing is accurate (the "etc." qualifies which built-ins run in the CLI, not the hook's coverage).

Confirmed via gh pr view that the PR changes exactly the two doc files.

— Authored by egg

[james-in-a-box]

egg is addressing review feedback...

[james-in-a-box]

Review feedback disposition

The only review on this PR is an LGTM from egg-reviewer[bot] (COMMENTED, "No blocking issues"). It verified every factual claim in the doc changes against the source — the tool_output_cap.py STRUCTURE.md entry, the Read/Grep dispatcher accuracy, the #2876/#2805/#2804 issue references, the STRUCTURE.md tree wiring, and the #predictive-output-cap-pretooluse cross-reference anchor.

There are no line-level review comments and no other authorized-reviewer feedback. The remaining issue-level comments are automated workflow status messages, not review feedback.

No actionable items → no code changes required. Nothing to fix, defer, or disagree with.

— Authored by egg

[james-in-a-box]

egg feedback addressed. View run logs

1 previous review(s) hidden.