This repo now houses two shapes of authored, version-controlled
content: long-form handbooks (retrieved via a KB) and short
redirect-skills (deployed into an agent's own skill directory).
A redirect-skill's entire job is pointing at a live or handbook
source instead of caching an answer that would go stale - the
gdext binding for Godot 4 is the first case, since it's pre-1.0
with no existing skill or MCP tool anywhere.
Neither shape is consumed directly from this repo: a handbook
isn't live until admitted to kb-engineering, a skill isn't live
until copied (never symlinked, since an agent may write back to
its own skill files) into wherever it actually runs.
Handbooks moved from the repo root into handbooks/, for symmetry
with skills/ - each content shape now owns its own directory
instead of handbooks implicitly owning the root.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011EBCcjJvEvhpToahqdsLLA
New repos only, same non-retroactive caveat as trunk naming.
Underscores don't survive the trip to a domain, a Docker tag, or
most package-registry names, where hyphens do; matches the
kebab-case already used for branch descriptions.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011EBCcjJvEvhpToahqdsLLA
Composes failure-handling.md (input contract discipline,
failure-propagation discipline) and adds a frontmatter convention
(title, collections) to both handbooks, fixing a collection-tag
drift found in the process.
Each handbook now opens with a flat YAML frontmatter block so the
two ingest-time values that actually describe the document -
title and collections - live with it instead of being retyped by
hand at admission time. That drift was real, not hypothetical:
the ingest recipe documented `--collection dev-practice` while the
actual admission used `handbooks`.
`version` deliberately stays out of frontmatter - a file can't
contain the hash of the commit that writes it - and remains a
value the ingest recipe computes fresh, now scoped per-file
(git log -1 -- <file>) rather than the whole repo's HEAD.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011EBCcjJvEvhpToahqdsLLA
Two sections, forged the same way as git-operations.md: input
contract discipline (degenerate input as the default case,
validate/sanitize/reject as distinct responses, the
meaning-preserving line for when sanitizing is safe) and
failure-propagation discipline (never substitute a plausible
artifact for a failed operation, retry/abort/surface, one
signaling convention per boundary, the anti-pattern catalogue).
Environment/verification parity was scoped out deliberately —
different axis (confidence-establishment, not code behavior),
earns its own handbook later rather than a section here.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011EBCcjJvEvhpToahqdsLLA
Fixes a factually reversed rebase conflict-marker explanation found
during review, plus four smaller accuracy/clarity corrections in the
same document.
The rebase conflict-marker explanation had HEAD and the incoming
side backwards, verified by reproducing all three merge/rebase
scenarios in a throwaway repo. Rewrote that passage around the
actual behavior instead of a merge-vs-rebase binary, since the old
"merge = HEAD is trunk" claim was also only true for one of the two
merge scenarios the doc itself describes.
Also fixes, from the same review pass:
- push-rejection handling in Branching split into its two real
causes (protection/permission vs. plain non-fast-forward)
- the commit-body trigger reworded to match how it's actually used
("why worth keeping", not just subject length)
- a mislabeled checkout equivalent for unstaging (that was always
git reset's job, not checkout's)
- the vendor-sync example no longer assumes upstream's trunk is
named master
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011EBCcjJvEvhpToahqdsLLA
Branching through tags, forged collaboratively over a full session:
trunk-based branching with type/description naming, atomic commits,
rebase-to-stay-current paired with merge --no-ff to land, revert vs.
reset vs. switch/restore split by shared-vs-local history, an
origin/upstream/mirror remote-naming discipline (including a
vendor-branch pattern for reconciling borrowed-repo drift), and
SemVer tags driven by the same type taxonomy as commits.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011EBCcjJvEvhpToahqdsLLA
mikkeli