fledge-plugin-standup

Generate a paste-ready Markdown standup post from your recent git commits, narrated by whatever LLM you've configured via fledge ai use. Works on a single repo, a list of repos, an entire dev directory, or every repo on GitHub you've touched.

Rewritten in Rust (v0.3.0+). The previous bash version needed python3 (date math) and jq (JSON parsing) at runtime; the Rust version handles both in-process. git is always required; gh is required for --gh.

$ fledge standup --gh --since "1 week ago"
## Yesterday
- Prepared v0.15.3 release including FLEDGE_PLUGIN_DIR support and plugin.toml as a version source (fledge)
- Re-absorbed templates-remote and doctor plugins back into core (fledge)
- Built standup and roast plugins from scratch (fledge-plugin-standup, fledge-plugin-roast)
- Hand-authored 33 spec stubs and tightened entitlements (Mono)

## Today
- (inferred) Cut v0.15.3 and watch the post-release formula PR land

Install

fledge plugins install CorvidLabs/fledge-plugin-standup

fledge plugins install auto-detects Cargo.toml and runs cargo build --release for you — only a Rust toolchain is required at install time.

Make sure an AI provider is configured:

fledge ai use                       # interactive picker
# or
fledge ai use ollama gpt-oss:120b-cloud

Modes

Single repo (default)

fledge standup                              # last 24h, everyone in the repo
fledge standup --me                         # just your commits (matches by email)
fledge standup --since "1 week ago" --me
fledge standup --author "Leif"              # someone else

--me matches by git config user.email first (stable across squash-merges), falling back to user.name.

Multiple repos

# Explicit list
fledge standup --repos ~/dev/foo,~/dev/bar,~/dev/fledge --me --since "1 week ago"

# Auto-discover one level deep
fledge standup --repo-dir ~/Development --me --since "1 week ago"

Output groups commits under ## <repo-name> headers. The LLM annotates each "Yesterday" bullet with the repo it belongs to: - Bumped tokei to library mode (fledge-plugin-metrics).

GitHub-wide

# Every commit you pushed anywhere on GitHub (requires `gh auth`)
fledge standup --gh --since "1 week ago"

# A specific GitHub user (e.g. you on a different machine, a teammate)
fledge standup --gh --gh-user 0xLeif --since yesterday

--gh shells out to gh search commits so it doesn't need local clones. Sees only commits you pushed to GitHub-visible repos. Requires gh (the JSON parsing and date math that used to need jq/python3 are now built in). Only commits visible to your authenticated user count, so this won't surface private-repo work you don't have access to.

Output controls

fledge standup --raw                        # plain aggregated log, skip the LLM
fledge standup --include-diff               # adds shortstat per commit (single-repo only)
fledge standup --show-prompt                # debug — prints what would be sent

Provider override

fledge standup --me -- --provider ollama --model qwen3-coder:480b-cloud

Anything after -- is forwarded to fledge ask.

How it works

A single Rust binary that:

1. Resolves the active mode (single / --repos / --repo-dir / --gh) and aggregates commits into a single log with ## <repo> headers when multi-repo. Each commit line carries its YYYY-MM-DD date so the model can split today vs. earlier.
2. Builds a structured prompt with two required sections (## Yesterday, ## Today)
3. Hands it off to fledge ask --no-spec-index "<prompt>" so it inherits your provider, model, timeouts, and rate-limit config

The model is told to:

• Put past-tense bullets for commits dated before today under "Yesterday" (stripping feat:/fix: prefixes)
• Put past-tense bullets for commits dated today under "Today"
• If there were no commits today, fall back to 1–3 inferred next-step bullets prefixed with (inferred)
• Annotate multi-repo bullets with the repo name in parentheses

Caveats

• --gh is bounded to 200 commits per call (the gh search commits --limit cap). For long windows, narrow with --gh-user or --since.
• --include-diff is single-repo only. Multi-repo diff stats would blow the token budget.
• Output quality scales with model — cloud models produce cleaner narration than local ones.
• Two standups for the same window produce structurally identical but stylistically different prose. That's fine.

License

MIT