bulk-reader
Answer a narrow question over one or more large repository files without reading their raw content into context.
Skill by threecee
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/threecee-bulk-reader
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: documents, writing
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/bulk-reader
What's inside
- Skill name
bulk-reader- Description
- Answer a narrow question over one or more large repository files without reading their raw content into context. Use when a bounded-read hook blocks an unbounded read of a file over the configured line threshold, or when the orchestrator needs facts (names, line numbers, counts, "does X exist") from a long spec, log, or generated document without spending context on its full text. Never for judgment calls.
- License
- MIT
- Instructions
- 57 lines in SKILL.md · view on GitHub
Files (1)
- SKILL.md
SKILL.md (first lines)
# Bulk Reader
Route a large-file read through the factory's bulk-read command instead of
loading the file into this session's context. The command sends the file(s)
plus one narrow question to the configured worker chain and returns
structured bullets only — never the raw corpus. The rules live in
`harness/bulk-read-contract.md`; this skill is the invocation.
## Invocation
```sh
{{BULK_READ_CMD}} <path> [<path> ...] --question "<question>"
```
- `<question>` must be answerable from the file's text: names, line numbers,
counts, "does X exist", "list every Y" — not "summarise this" or an
open-ended judgment (contract §3 rule 2).
- Repeat the path argument for more than one input; every file is streamed
to the worker in one request, not read into this session.
- Paths must be inside the repository or the lane scratchpad. A path outside
that boundary is refused with exit 2 before any worker is called; pass
`{{ALLOW_EXTERNAL_FLAG}}` only after confirming the material is approved
for the worker's provider under the project's data rules (contract §3
rule 3). A file path never implies approval.
- Exit 3 means every configured worker failed; nothing from the file reaches
stdout or stderr (contract §3 rule 6). Fall back to a bounded read.
- Every bullet leads with the exact identifier. Before you act on one, read
the cited line in the source with a bounded read (contract §3 rule 4).
## When NOT to use this
Never for debugging, root cause, architecture judgment, boundary-sensitive
code, or security review — the worker answers the literal question against
the literal text with no memory of this session's goals. Read the file
directly when the judgment itself needs the full text, not facts extracted
from it (contract §9).
## Escape hatch
A targeted read with `offset`/`limit` (or a narrow `sed -n` span, `grep`,
`wc`) always passes the hook unchanged — use that instead of this skill when
only a known slice of the file is needed. `{{SHUNT_ALLOW}}` exempts named
paths. `{{SHUNT_DISABLE}}` turns the bounded-read hooks off for the session
(for example when every configured worker is down); it is a session-wide
bypass, not a per-call override — unset it once the workers recover
(contract §3 rule 5, §5).
## Parameters
`{{BULK_READ_CMD}}`, `{{ALLOW_EXTERNAL_FLAG}}`, `{{SHUNT_ALLOW}}`,
`{{SHUNT_DISABLE}}` and the line threshold are bound in the repository's
operations doc from the table in `harness/bulk-read-contract.md` §4. The
source factory bound them to `python scripts/factory/bulk_read.py`,
`--allow-external`, `SHUNT_ALLOW`, `SHUNT_DISABLED=1` and 350 lines; those
are one operator's choices, not defaults of this skill.From threecee/software-factory-kit, MIT.
More from threecee/software-factory-kit
- 0.563score
compare-screenshots
threecee
Compare screenshots to judge which image is less wrong, not to match a baseline.
SkillBFreshness: B★ 2Claude Code +9 - 0.563score
refactor-clean
threecee
Refactor toward one clear owner per concept instead of adding sediment.
SkillBFreshness: B★ 2Claude Code +9 - 0.563score
code-review
threecee
Review a current diff for correctness, stale references, ownership, test quality, unnecessary compatibility, and cleanup.
SkillBFreshness: B★ 2Claude Code +9 - 0.559score
brainstorming
threecee
Explore intent and produce an approved design before unbriefed creative implementation.
SkillBFreshness: B★ 2Claude Code +9 - 0.559score
finishing-a-development-branch
threecee
Use after implementation and tests are complete to choose and safely execute merge, pull-request, keep, or discard handling while respecting normal, linked…
SkillBFreshness: B★ 2Claude Code +9 - 0.559score
independent-lane-review
threecee
Run a fresh-context, multi-model adversarial review of a lane before train assembly or on demand.
SkillBFreshness: B★ 2Claude Code +9 - 0.559score
implement-spec
threecee
Implement an approved spec through reviewable passes.
SkillBFreshness: B★ 2Claude Code +9 - 0.559score
audit-choices
threecee
Audit decisions an implementing agent made beyond the brief, without changing code.
SkillBFreshness: B★ 2Claude Code +9 - 0.557score
handback-digest
threecee
Draft one choices-ledger entry per choice in a lane handback, in the ledger's own ELI5 entry format, as the starting point for the orchestrator's…
SkillBFreshness: B★ 2Claude Code +9 - 0.557score
claude
threecee
Use Claude Code as an independent `claude -p` subagent when the user explicitly asks for Claude, wants a second-agent opinion from Claude, or asks to delegate…
SkillBFreshness: B★ 2Claude Code +9 - 0.557score
log-triage
threecee
Extract failing tests, assertion messages, tracebacks, and red gate results from a large run or verify log without reading its full text.
SkillBFreshness: B★ 2Claude Code +9
Signals
- Stars
- 2
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Sep 6, 2026 (1 month ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/threecee/software-factory-kit
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from threecee
- 0.563score
compare-screenshots
threecee
Compare screenshots to judge which image is less wrong, not to match a baseline.
SkillBFreshness: B★ 2Claude Code +9 - 0.563score
refactor-clean
threecee
Refactor toward one clear owner per concept instead of adding sediment.
SkillBFreshness: B★ 2Claude Code +9 - 0.563score
code-review
threecee
Review a current diff for correctness, stale references, ownership, test quality, unnecessary compatibility, and cleanup.
SkillBFreshness: B★ 2Claude Code +9 - 0.559score
brainstorming
threecee
Explore intent and produce an approved design before unbriefed creative implementation.
SkillBFreshness: B★ 2Claude Code +9 - 0.559score
finishing-a-development-branch
threecee
Use after implementation and tests are complete to choose and safely execute merge, pull-request, keep, or discard handling while respecting normal, linked…
SkillBFreshness: B★ 2Claude Code +9 - 0.559score
independent-lane-review
threecee
Run a fresh-context, multi-model adversarial review of a lane before train assembly or on demand.
SkillBFreshness: B★ 2Claude Code +9
Cite this page
bulk-reader skill. Agent Mart. https://agentmart.dev/skills/threecee-bulk-reader (last updated 2026-10-24)