v0.2.32

feat(plan): typed block markers, `plan unblock`, and an autonomous re…

v0.2.31

feat(telegram): read-only notifier pilot (spec status + plan run) (#53)

Add `csdd telegram {init,run}`, a read-only, outbound-only bridge that relays plan-run progress and spec-status changes to a Telegram chat — the `csdd web` "view, never author" contract, delivered as messages.

- `telegram init` saves the bot token + chat_id to `.csdd/bot.json` (0600, gitignored) from flags/env/prompt, then sends a test message.
- `telegram run` polls two feeds and relays deltas: new lines in each plan's run journal (docs/plans/<slug>/log.md) and changes to any spec's phase/approvals (specs/<feature>/spec.json). Startup seeds baselines so history is never replayed.
- No new dependency (net/http); TELEGRAM_API_BASE overrides the endpoint.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N3FKDAkdkVhhUiV8gPWDQo

v0.2.30

fix(graph): close correctness, robustness, and consumption gaps in th…

v0.2.29

refactor(plan): plan run drops all git; session's csdd dev-cycle owns…

v0.2.28

feat(plan): plan run always bypass-mode with sandbox consent gate; ap…

v0.2.27

feat: devcontainer Go + sandbox csdd; hooks/pre-push opt-in (+ Java) …

v0.2.26

feat(plan): plan mode, decision records, and the glossary (#48)

* feat(plan): plan mode, decision records, and the glossary

Land the plan layer above specs and the two knowledge contracts that ride on it.

Plan mode (substrate): docs/plans/<slug>/ decomposes an initiative into feats,
each becoming exactly one spec. Adds the plan grammar (parse/validate), the
approval hash gate, the sequencer, deterministic briefs, the autonomous runner
and its sandbox, the /prd + quick-prd authoring skills, plan/feat graph nodes,
and the web read model.

Decision records (docs/adr/): a decision passing the triple gate — hard to
reverse, surprising without context, a real trade-off — becomes an ADR at
docs/adr/NNNN-<slug>.md. Feats cite it as adr:<slug>, a third ref token beside
stack: and [[wiki]]. `csdd plan validate` reports broken/ambiguous/malformed and
cites-superseded citations plus docs/adr well-formedness; `csdd plan brief`
inlines cited records in full with the open-decision forbidden line; the graph
emits adr nodes with cites/superseded_by edges and a broken/orphan-decision lint.
The prd skill grills gate-positive decisions one at a time (recording each ADR
inline), sweeps assumptions at Present, and funnels run deviations through
supersession; quick-prd carries the light gate. Dogfood: ADRs 0001-0003.

The glossary (docs/glossary.md): one canonical term per domain concept with an
_Avoid_ synonym list. A shared internal/glossary package parses, normalizes, and
whole-token matches identifiers. `csdd plan validate` reports an avoided term in a
feat/plan slug; `csdd graph analyze` / `csdd wiki lint` catch spec dirs and wiki
page names; the graph emits term nodes and references edges (orphan-term lint).
The glossary skill (challenge/sharpen/stress-test/record, the tombstone rule) and
/glossary command own the discipline; prd/quick-prd/wiki gain invocation hooks.
Dogfood: docs/glossary.md.

* fix(plan): satisfy golangci-lint and the Windows test runner

- runner.go: check the fmt.Fprintf return in the log closures and summarize
  (errcheck).
- sandbox_test.go: skip the POSIX exec-bit assertion on Windows, which has no
  exec bit (matches the web_test.go token-perm guard).
- glossary_cli_test.go: drop the ineffectual code assignment in the analyze step.

v0.2.25

feat(graph): knowledge base — deterministic csdd graph + LLM-authored…

v0.2.24

feat: agents/skills workflow, agent colors, and CLI/tooling hardening…

v0.2.23

fix(cli): drop duplicate copy registration from the #39/#40 merge (#41)

PRs #39 and #40 both added the `csdd copy` command; squash-merging #40 on
top of #39 left two `case "copy"` arms in the dispatcher — a duplicate-case
compile error that broke `go build` on main — plus a redundant RESOURCES
help line and a duplicate copy example in the shipped CLAUDE.md cheat sheet.
Remove the duplicates, keeping #39's canonical wording. Build, vet, gofmt,
and the full test suite are green.


Claude-Session: https://claude.ai/code/session_01Xg9KzFKN8bbaL5xRexm3YE

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>