You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
translation: refuse remote or translated --from, lower-case slug tag (C67-C69)
set: --unpin, named pin errors, review_by date and self-exception checks; audit lists overdue exceptions
evals: check consumers before changing a shared decision
graph: consumers lists documents here and in remotes that name a shared document
check: review_by and pins are common keys; document T5 as error
pins: pin remote documents by body revision, rule L7, spec set --pin (X2)
exceptions: rel.exception_of with review_by, rule T5 (X4)
remotes: L6 cites, R1 for remote sources, set --source-rev, doctor lists remotes; translation evals
translations: validate lang and source_rev, reject translation chains, warn on duplicate languages, schema keys, census sections (C57-C61)
Name documents in other repositories as : through [remotes]
Pair translations by place: [translations] layout, rule R2, spec new --translation, spec fix --translations
Quiet config warnings outside check/doctor, open a document by its full timestamp, write imported relations as ids, name changed sections of translations, census --fail-on
translations: rel.translation_of, source_rev, rule R1, census --translations
check: A4 skips the CI runner home, A3 skips git refs
import: presets read both spellings, link superseding records, rewrite links, name the next fix
check: S7 required sections per type ([types.] sections)
check: A4 machine-specific home paths in agent assets
cli: check resolves names like resolve, list/ls find, near-miss status and path candidates, fix and show next lines; docs: H1 in rules reference
check: H1 reports invisible characters in documents and agent assets, with a removal fix
lsp help: spec.run acts on disk; every change re-runs the full check
lsp: one quick fix per candidate; offered --write fix commands run through workspace/executeCommand
check: name the naming mismatch for unsaved files, mention --stdin-path only when used, explain '-'; lsp: -32700 for bad JSON and a reason on stderr for broken frames; binstall metadata
check: accept document ids; clearer error for unknown names
lsp: keep the last full check and reuse it until disk or buffers change
lsp: offer replacement fixes as quick fix code actions
doctor: no spec.toml is a finding; check: next for errors left out by --fail-on; new: full rerun with --slug
lsp: diagnostics over stdio; --stdin-path accepts a new file in a spec directory
check: --stdin-path for unsaved buffers; clear errors for files that are not documents
check: report one file's findings and an editor format; mv --cited-paths reads tracked files
Cycle 22: deterministic doctor/guard tests, generated-file guidance, large-corpus regression test
Cycle 21: README standardization, llms.txt, doctor phase 2, show splitting, mv --tier, S3/guard hints
feat: --max-bytes list budget (MCP default 40000), graph --assets, spec doctor
graph impact lists the agent assets that name the document or what reaches it
find --open-tasks: documents with unchecked task items, with the count
Revert "Merge feat/cycle18"
Check published count markers: count_paths for README and docs, with a test
Cycle 17: positional title, fail-on none and excluded-error summary, real ids in next, Action version from its own tag, slug next
docs: byte-identical twins (md-kit v1)
docs: v5 information architecture, md-kit components and touch targets
Number repeated fingerprints; action gains fail-on and checks a pull request from its merge base
Add a composite GitHub Action that installs a verified release and runs spec check
Add check --format github, sarif, gitlab and rdjson; pre-commit hooks use language unsupported
Errors carry one runnable next command; missing args and did-you-mean in usage errors; set refuses superseded without a successor and warns unknown keys; non-ASCII titles slug from governs; first --local excludes the local dir from git; graph impact on a path suggests governs; real rule ids in next
help --markdown prints a reference page and suggests the nearest name; mv and import refuse [forbid] paths; set no longer guesses a finishing status; init shows a commented count_paths and names spec explain
new refuses [forbid] paths and suggests --local; set names reachable statuses and a next command; N1 reads [check] count_paths; help --markdown keeps a blank line before the table
docs: fix site layout overflow and the commands table; list the install channels
check: N1 compares numbers in spec:count markers with find filters or table rows; spec fix rewrites them
graph: --format box draws layered boxes sized to COLUMNS, ASCII on request or non-UTF-8 locale, tree fallback; tests pin the locale
graph: tree with relation and distance, impact relations, chain both ways, governs in json, mermaid/dot, --ascii/--depth
Status moves name a real way out: a draft is activated before it is superseded; S4 names the status meant for accepted, proposed, deprecated
Built-in change type: one user-facing change per note, active until a release archives it with release set (N8)
docs: build the site with Jekyll on GitHub Pages
spec sync assets: copy each [[assets]] copy_of source over its root; preview until --write, --prune to delete, refuses paths outside the root and symbolic links (A8)
[check] extra_keys: frontmatter keys every type may carry, prefix* allowed (S1)
docs: set the Pages custom domain (CNAME)
verified_at pin: spec set resolves HEAD, audit stale reports commits to governed code since the pin
assets name documents: docs list in spec assets, assets field in spec show
docs: short README with generated command table, docs/ guides, and drift tests
spec agents: name edit, check summary, assets and audit kinds; commit hash in --version
feat(edit): --row with --cell COLUMN=VALUE or a whole row; --append adds a missing row
feat(new): spec new skill/rule/agent writes id-less templates
feat(audit): drift kind for skill copies that differ from .agents/ and CLAUDE.md without @AGENTS.md
feat(check): A3 paths and spec commands named in agent assets exist
feat(check): A1 asset frontmatter and A2 unique asset names
feat(assets): declare agent asset roots in spec.toml and list them with spec assets
feat(edit): add --new-section with --after placement
feat(audit): report plans whose tasks and status disagree (kind finished)
feat(edit): read piped or redirected stdin when neither --text nor --file is given; add --replace
feat(check): S1 unknown key names a near key or the spec.toml fields line that declares it
fix(edit): rerun hint uses --text= so text starting with '-' still parses
feat(dates): write created/updated as RFC 3339 UTC, read plain dates, offsets and UTC/GMT/KST/JST
docs: say that timestamp keys and dates are UTC
feat(edit): rerun hint quotes the input so it can be pasted
feat(fix): resolve a shared alias (I2) by keeping it on the picked document
feat(check): --summary counts per rule, -q lists errors only, clearer paging and baseline lines
feat(check): --new lists documents not in the last commit; say audit is optional
feat(links): match stale path cites by file name and rewrite them on mv
feat(edit): replace or append to a section or the whole body
feat(show): count task items and list them with --tasks