cngkit 1.1.2 → 1.1.4

package/LICENSE

@@ -4,4 +4,4 @@ This software is provided as personal, opinionated Curly.ng tooling.
 
 You can use it, modify it, and share it. If it helps you, support the workshop.
 
-Support: https://curly.ng/donate
+Support: https://curly.ng/support

package/README.md

@@ -13,15 +13,18 @@ the Cloudflare-backed Harness knowledges catalog:
 - `cngkit coderoom share [room-code]` starts a live shared code room from the current directory.
 - `cngkit coderoom join <room-code>` joins another developer's shared code room.
 - `cngkit scrub [path]` scans a file or directory with TruffleHog and prints a redacted report.
-- `cngkit scrub [path] --yes` rewrites detected secret values inline with `CNGKIT_SECRET` placeholders.
+- `cngkit scrub [path] --yes` rewrites detected secret values inline with `[CNGKIT_SECRET:<detector>:<verified|unverified>]` placeholders.
+- `cngkit transcripts list [--source all|codex|claude] [--limit n]` lists recent local Claude/Codex transcript files.
+- `cngkit transcripts read <path-or-session-id> [--limit n]` prints normalized user/assistant transcript entries.
+- `cngkit transcripts grep <query> [--limit n] [--file-limit n]` searches recent local transcript entries.
 - `cngkit knowledges status` prints the remote catalog state.
 - `cngkit knowledges audiences` lists available audience filters.
-- `cngkit knowledges search <query>` runs semantic search against Cloudflare Vectorize.
-- `cngkit knowledges list [query]` lists known subskills.
-- `cngkit knowledges files [query]` lists uploaded catalog files.
-- `cngkit knowledges read <file-path>` reads a catalog file excerpt.
-- `cngkit knowledges grep <pattern>` searches catalog file contents.
-- `cngkit knowledges glob [pattern]` lists catalog files by supported glob pattern.
+- `cngkit knowledges search <query> [--limit n]` runs semantic search against Cloudflare Vectorize (default limit 5).
+- `cngkit knowledges list [query] [--limit n]` lists known subskills (default limit 25).
+- `cngkit knowledges files [query] [--audience id] [--limit n]` lists uploaded catalog files (default limit 25).
+- `cngkit knowledges read <file-path> [--offset n] [--limit n]` reads a catalog file excerpt (default limit 200).
+- `cngkit knowledges grep <pattern> [--path path] [--include glob] [--output-mode content|files_with_matches|count] [--context n] [--case-insensitive]` searches catalog file contents (default mode `content`, default path `/`).
+- `cngkit knowledges glob [pattern] [--path path]` lists catalog files by supported glob pattern (default pattern `**/*.md`, default path `/`).
 - `.git/` and files ignored by the repo's `.gitignore` are not synced.
 - Later changes override earlier changes for the MVP conflict rule.
 
@@ -30,16 +33,28 @@ Help is baked into the CLI as Markdown with progressive disclosure:
 ```bash
 cngkit --help
 cngkit scrub --help
+cngkit transcripts --help
 cngkit knowledges --help
 cngkit knowledges read --help
 cngkit knowledges grep --help
 cngkit knowledges glob --help
 ```
 
+Global options accepted by every command:
+
+- `--api-base-url <url>` overrides the Curly backend URL (default: `CNGKIT_API_BASE_URL` or `https://curly.ng`).
+- `--version` prints the installed `cngkit` version.
+- `--help` / `-h` prints the progressive help for the current command.
+
 The canonical CLI spec lives at `docs/superpowers/specs/2026-06-25-cngkit-cli-spec.md`
 in the repository. The baked help source lives in `src/help-specs.ts`; keep both
 aligned when changing command behavior.
 
+Command routing is implemented with Pastel route files under `src/commands/`, rendered
+through Ink. `tsup` emits a file-preserving ESM build so the published package contains
+`dist/cli.js` plus `dist/commands/**`; the private workspace client stays bundled while
+public runtime packages remain normal package dependencies.
+
 `scrub` requires the `trufflehog` binary on `PATH`. On macOS, install it with:
 
 ```bash
@@ -48,6 +63,21 @@ brew install trufflehog
 
 Inline masking is intentionally gated behind `--yes` because it rewrites files in place.
 
+## Local Agent Transcripts
+
+The `transcripts` command reads local JSONL logs from `~/.codex/sessions`,
+`~/.codex/archived_sessions`, `~/.claude/projects`, and `~/.claude/history.jsonl`.
+It is local-only operator tooling: transcript contents are not uploaded to the Curly backend.
+
+```bash
+cngkit transcripts list --limit 12
+cngkit transcripts read <path-or-session-id> --source codex --limit 80
+cngkit transcripts grep "deploy failed" --source all --file-limit 60 --limit 20
+```
+
+By default, `read` and `grep` print only user/assistant text and skip internal prompt,
+hook, and tool noise. Use `--include-internal` when debugging transcript plumbing.
+
 ## Coderoom
 
 Coderoom is for quickly uniting developers around one working tree. One developer starts
@@ -80,12 +110,19 @@ cngkit knowledges glob "**/*.md" --path /libraries/lib-cloudflare
 ```
 
 Add `--json` to any `knowledges` command when another tool should consume the
-raw API response. `grep` also accepts `--output-mode <content|files_with_matches|count>`,
-`--include <glob>`, `--context <n>`, and `--case-insensitive`.
+raw API response. `grep` also accepts `--output-mode <content|files_with_matches|count>`
+(default `content`), `--include <glob>` (default `*`), `--context <n>` (default `0`), and
+`--case-insensitive`. Result limits default to `5` for `search` and `25` for `list` and
+`files`; `read` defaults to `200` lines with a backend cap of `2000`, and accepts `--offset`
+for paging.
+
+`read` and `grep` accept shorthand catalog paths: `/libraries/lib-cloudflare/SUBSKILL.md`
+expands to `skills/knowledges/subskills/libraries/lib-cloudflare/SUBSKILL.md`, mirroring the
+Claude `read`/`grep` tool convention.
 
 This package is mostly personal tooling for Curly.ng. It is intentionally opinionated,
 small, and practical rather than a general synchronization platform.
 
 ## License
 
-Beerware. If this helps you, support the workshop: https://curly.ng/donate
+Beerware. If this helps you, support the workshop: https://curly.ng/support