dev-digest-ai-marketplace

Shared engineering practices: done gate, spec-tier guide, module insights, review bundle and the agent guard script.

Category: engineering

Requires Claude Code >=2.1.110

What's new in 1.0.1 scripts/agent-guard.py: read-only agents could run an arbitrary program through git grep -O, either with the flag clustered into other short options (-nOsh) or with an abbreviated long option (--open-files-in-p). Abbreviated forms of --no-index, --ext-diff, --textconv and --output also slipped past. The guard now denies every short cluster containing O and every long option that abbreviates a denied flag.

No dependencies

/plugin marketplace add HlibHav/dev-digest-ai-marketplace
/plugin install engineering-paved-path@dev-digest-ai-marketplace

Read-only researcher agent that answers a concrete question from the repo or the web with evidence.

Category: research

Requires Claude Code >=2.1.110

What's new in 1.0.1 researcher combined WebFetch with NotebookLM write tools, so a prompt injection in a fetched page could make it copy repository content into a notebook. Removed notebook_create, notebook_rename, source_add, source_rename, note and tag from its tools list. It keeps listing, querying, reading sources and the research flow (research_start, which creates the notebook for new research, research_status, research_import).

No dependencies

/plugin marketplace add HlibHav/dev-digest-ai-marketplace
/plugin install research-tools@dev-digest-ai-marketplace

Independent read-only architecture reviewer that checks a diff against layer boundaries.

Category: review

Requires Claude Code >=2.1.110

What's new in 1.0.1 scripts/agent-guard.py: read-only agents could run an arbitrary program through git grep -O, either with the flag clustered into other short options (-nOsh) or with an abbreviated long option (--open-files-in-p). Abbreviated forms of --no-index, --ext-diff, --textconv and --output also slipped past. The guard now denies every short cluster containing O and every long option that abbreviates a denied flag.

/plugin marketplace add HlibHav/dev-digest-ai-marketplace
/plugin install architecture-review@dev-digest-ai-marketplace

Spec-driven workflow: idea to spec to plan, delegating research and architecture review.

Category: workflow

Requires Claude Code >=2.1.110

What's new in 1.1.1 workflow-retro deep --since <run start> reported no multi-agent run when the run happened in another session. Without --session, scripts/collect.py read only the newest transcript in the project, which is the retro's own session. With --since and no --session it now scans every transcript in the project with lines at or after --since and reports each session that launched agents, skipping the ones that didn't. If none is found it exits with "no multi-agent run found". --session, and a run with neither flag, behave as before. A --since without a timezone is now read as UTC; before, it crashed …

/plugin marketplace add HlibHav/dev-digest-ai-marketplace
/plugin install sdd-engineering@dev-digest-ai-marketplace