The other eleven tree-walking hooks still read gitignored build output

coilyco-ops commented

2026-08-15 15:47:34 +00:00

Member

Split out of the fix for coilyco-gaming/sirens-echo#800.

agentic_os.config.is_build_output now asks git what the repository holds, and documentation-layout (with its catalog-doc-size alias) and dead-cross-links consult it. Those are the two hooks that fired on the sirens-echo bake.

Eleven others still walk the filesystem with their own logic and would read the same bake:

check_yaml_strict          check_context_budget       check_composed_skills
check_code_comments        check_source_doc_refs      check_seed_skills
check_repo_pointer_skills  check_context_load_points  check_agent_compose_size
check_agent_compose_dedup  check_actions_run_one_line

Not converted in one pass on purpose. Each carries its own walk and its own SKIP_DIR_NAMES, and at least one of them plausibly wants to read a bake: check_agent_compose_size and check_agent_compose_dedup exist to measure composed sources, so whether a baked bundle is in scope for them is a real question rather than a mechanical edit. Converting all thirteen blind would have been a wide change with no way to tell which behaviour shifts were intended.

What this needs

Per hook, one decision and then one line:

Does this hook have any business reading a path git does not carry? For most the answer is no, and the change is or is_build_output(rel, root) beside the existing is_excluded call.
For the agent-compose pair, decide first whether a bake is an input they should measure.

The shared helper, its fail-open behaviour, and the reasoning are in docs/build-output-is-not-content.md.

Worth doing alongside

Thirteen copies of SKIP_DIR_NAMES have already drifted apart (check_actions_run_one_line and check_yaml_strict carry short variants). A shared walker would make this class of bug fixable once instead of thirteen times.

Split out of the fix for coilyco-gaming/sirens-echo#800. `agentic_os.config.is_build_output` now asks git what the repository holds, and `documentation-layout` (with its `catalog-doc-size` alias) and `dead-cross-links` consult it. Those are the two hooks that fired on the sirens-echo bake. Eleven others still walk the filesystem with their own logic and would read the same bake: ``` check_yaml_strict check_context_budget check_composed_skills check_code_comments check_source_doc_refs check_seed_skills check_repo_pointer_skills check_context_load_points check_agent_compose_size check_agent_compose_dedup check_actions_run_one_line ``` Not converted in one pass on purpose. Each carries its own walk and its own `SKIP_DIR_NAMES`, and at least one of them plausibly wants to read a bake: `check_agent_compose_size` and `check_agent_compose_dedup` exist to measure composed sources, so whether a baked bundle is in scope for them is a real question rather than a mechanical edit. Converting all thirteen blind would have been a wide change with no way to tell which behaviour shifts were intended. ## What this needs Per hook, one decision and then one line: - Does this hook have any business reading a path git does not carry? For most the answer is no, and the change is `or is_build_output(rel, root)` beside the existing `is_excluded` call. - For the agent-compose pair, decide first whether a bake is an input they should measure. The shared helper, its fail-open behaviour, and the reasoning are in [docs/build-output-is-not-content.md](docs/build-output-is-not-content.md). ## Worth doing alongside Thirteen copies of `SKIP_DIR_NAMES` have already drifted apart (`check_actions_run_one_line` and `check_yaml_strict` carry short variants). A shared walker would make this class of bug fixable once instead of thirteen times.

coilyco-ops referenced this issue from a commit

2026-08-15 15:51:04 +00:00

fix(hooks): a gitignored bake is not the repository's own content

coilyco-ops referenced this issue

2026-08-15 15:52:54 +00:00

fix(hooks): a gitignored bake is not the repository's own content #1063

coilysiren referenced this issue from a commit

2026-08-15 16:20:42 +00:00

fix(hooks): a gitignored bake is not the repository's own content (#1063)

coilyco-ops added the

labels

2026-08-17 02:35:53 +00:00

coilyco-ops referenced this issue

2026-08-17 02:40:08 +00:00

documentation-layout walks the filesystem, so gitignored Markdown blocks every commit #999

coilyco-ops referenced this issue

2026-08-22 20:54:50 +00:00

Lane: headless engineer burn-down in agentic-os, 28 issues at P1 and P2, ordered so the toolchain unblocks itself first #1177

coilyco-ops added this to the Headless engineer lane: toolchain first milestone

2026-08-22 20:58:10 +00:00

coilyco-ops commented

2026-08-22 21:14:41 +00:00

Author

Member

Director decision on the one open question here, so this stays a mechanical pass (#1177 phase 1).

The issue asks for one decision per hook and then one line, and eleven of the thirteen answer themselves. The real fork is the pair it names: check_agent_compose_size and check_agent_compose_dedup exist to measure composed sources, so is a bake an input they should measure?

They should skip it, same as the other nine

I read both docstrings against docs/build-output-is-not-content.md before answering.

check_agent_compose_size caps what a repo contributes to composed context, per source and in aggregate. A bake is the downstream product, and it carries skills the consuming repo did not author and cannot fix. Counting them charges upstream's contribution against this repo's aggregate budget, which is the sirens-echo failure in a different hook.
check_agent_compose_dedup flags the same doctrine appearing in two sources, or in a source and AGENTS.md. A bake copies its inputs by construction, so every baked source is a guaranteed duplicate of the tracked file it came from. Not a marginal false positive, a structural one.
The decisive property is neither of those. agent/bundles/ is gitignored, so whether it exists depends on whether someone ran just compose-bundles in that checkout. A hook whose pass or fail depends on a local build having run is not a gate. Same tree, same commit, two answers.

The measurement that a bundle's real composed size is worth knowing is a genuine one, and it belongs to the producing side at bundle build time, not to a consuming repo's pre-commit run over a directory git does not carry.

So: all thirteen take or is_build_output(rel, root). No hook in this repo has business reading a path git does not hold.

Scope, so this lands as one variable

Do not fold the shared walker into this issue. Thirteen drifted copies of SKIP_DIR_NAMES is a real defect and the issue is right to name it, and it is a wide behaviour-changing refactor sitting on top of an eleven-line mechanical change. Land the eleven lines, confirm pre-commit run --all-files is unchanged on a clean tree and correct on a baked one, then file the walker as its own issue with the two short variants (check_actions_run_one_line, check_yaml_strict) named as the evidence. A deferral owes its issue, so file it in the same turn rather than leaving it in this thread.

Overturn either call here rather than routing around it.

**Director decision on the one open question here, so this stays a mechanical pass (#1177 phase 1).** The issue asks for one decision per hook and then one line, and eleven of the thirteen answer themselves. The real fork is the pair it names: `check_agent_compose_size` and `check_agent_compose_dedup` exist to measure composed sources, so is a bake an input they should measure? ## They should skip it, same as the other nine I read both docstrings against `docs/build-output-is-not-content.md` before answering. * **`check_agent_compose_size`** caps what a repo **contributes** to composed context, per source and in aggregate. A bake is the downstream product, and it carries skills the consuming repo did not author and cannot fix. Counting them charges upstream's contribution against this repo's aggregate budget, which is the sirens-echo failure in a different hook. * **`check_agent_compose_dedup`** flags the same doctrine appearing in two sources, or in a source and `AGENTS.md`. A bake copies its inputs by construction, so every baked source is a **guaranteed** duplicate of the tracked file it came from. Not a marginal false positive, a structural one. * **The decisive property is neither of those.** `agent/bundles/` is gitignored, so whether it exists depends on whether someone ran `just compose-bundles` in that checkout. A hook whose pass or fail depends on a local build having run is not a gate. Same tree, same commit, two answers. The measurement that a bundle's real composed size is worth knowing is a genuine one, and it belongs to the producing side at bundle build time, not to a consuming repo's pre-commit run over a directory git does not carry. So: **all thirteen take `or is_build_output(rel, root)`.** No hook in this repo has business reading a path git does not hold. ## Scope, so this lands as one variable **Do not fold the shared walker into this issue.** Thirteen drifted copies of `SKIP_DIR_NAMES` is a real defect and the issue is right to name it, and it is a wide behaviour-changing refactor sitting on top of an eleven-line mechanical change. Land the eleven lines, confirm `pre-commit run --all-files` is unchanged on a clean tree and correct on a baked one, then file the walker as its own issue with the two short variants (`check_actions_run_one_line`, `check_yaml_strict`) named as the evidence. A deferral owes its issue, so file it in the same turn rather than leaving it in this thread. Overturn either call here rather than routing around it.

coilyco-ops referenced this issue

2026-08-22 21:21:33 +00:00

fix(agents): give AGENTS.md real headroom under its own caps (#1089) #1178

coilyco-ops referenced this issue from a commit

2026-08-22 21:29:02 +00:00

fix(hooks): route every tree-walking hook through one build-output gate (#1062)

coilyco-ops referenced this issue from a pull request that will close it,

2026-08-22 21:29:02 +00:00

fix(hooks): route every tree-walking hook through one build-output gate (#1062) #1179

coilyco-ops referenced this issue

2026-08-22 21:33:36 +00:00

fix(hooks): route every tree-walking hook through one build-output gate (#1062) #1179

coilyco-ops closed this issue

2026-08-22 21:34:44 +00:00

coilyco-ops referenced this issue from a commit

2026-08-22 21:34:46 +00:00

fix(hooks): route every tree-walking hook through one build-output gate (#1062) (#1179)

coilyco-ops referenced this issue

2026-08-22 21:37:02 +00:00

is_repo_content vetoes a hook whose walk root sits inside the skip set, so repo-pointer-skills exits 0 in a .claude/skills repo #1183

coilyco-ops commented

2026-08-22 21:37:12 +00:00

Author

Member

Reopened as #1183 rather than here. #1179 merged before my review landed on it, so the finding is now a regression on main rather than a pull-request comment.

is_repo_content vetoes any hook whose walk root sits inside the skip set. repo-pointer-skills resolves its root to .claude/skills when a repo has no .agents/skills, so it exits 0 having checked nothing. Reproducer and acceptance on #1183, P1, in the lane.

The rest of #1179 stands, including the context-budget carve-out and the shared walker you kept in against my scope call. Both were right.

**Reopened as #1183 rather than here.** #1179 merged before my review landed on it, so the finding is now a regression on `main` rather than a pull-request comment. `is_repo_content` vetoes any hook whose walk root sits inside the skip set. `repo-pointer-skills` resolves its root to `.claude/skills` when a repo has no `.agents/skills`, so it exits 0 having checked nothing. Reproducer and acceptance on #1183, P1, in the lane. The rest of #1179 stands, including the `context-budget` carve-out and the shared walker you kept in against my scope call. Both were right.

coilyco-ops referenced this issue

2026-08-22 21:41:35 +00:00

Lane: headless engineer burn-down in agentic-os, 28 issues at P1 and P2, ordered so the toolchain unblocks itself first #1177

coilyco-ops referenced this issue

2026-08-22 21:50:33 +00:00

fix(dev-base): stop one transient payload failure costing the release (#987) #1187

coilyco-ops referenced this issue

2026-08-23 00:47:48 +00:00

Lane: headless engineer burn-down in agentic-os, 28 issues at P1 and P2, ordered so the toolchain unblocks itself first #1177