rh-cli 0.1.0__tar.gz → 0.2.1__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- rh_cli-0.2.1/CHANGELOG.md +512 -0
- rh_cli-0.2.1/MANIFEST.in +7 -0
- rh_cli-0.2.1/PKG-INFO +287 -0
- rh_cli-0.2.1/README.md +257 -0
- rh_cli-0.2.1/ROADMAP.md +236 -0
- rh_cli-0.2.1/examples/README.md +72 -0
- rh_cli-0.2.1/examples/configure-env.sh +28 -0
- rh_cli-0.2.1/examples/envs/README.md +19 -0
- rh_cli-0.2.1/examples/envs/claude.sh +15 -0
- rh_cli-0.2.1/examples/envs/codex.sh +15 -0
- rh_cli-0.2.1/examples/envs/pi.sh +19 -0
- rh_cli-0.2.1/examples/prompts/critique.md +21 -0
- rh_cli-0.2.1/examples/prompts/literature.md +29 -0
- rh_cli-0.2.1/examples/prompts/orchestrator.md +27 -0
- rh_cli-0.2.1/examples/prompts/researcher.md +24 -0
- rh_cli-0.2.1/examples/prompts/reviewer.md +24 -0
- rh_cli-0.2.1/examples/prompts/writer.md +19 -0
- rh_cli-0.2.1/examples/spawn-critique.sh +28 -0
- rh_cli-0.2.1/examples/spawn-researcher.sh +38 -0
- {rh_cli-0.1.0 → rh_cli-0.2.1}/pyproject.toml +8 -8
- {rh_cli-0.1.0 → rh_cli-0.2.1}/src/rh/__init__.py +4 -4
- {rh_cli-0.1.0 → rh_cli-0.2.1}/src/rh/add.py +2 -2
- rh_cli-0.2.1/src/rh/agents.py +280 -0
- rh_cli-0.2.1/src/rh/artifacts.py +279 -0
- rh_cli-0.2.1/src/rh/checklist.py +584 -0
- rh_cli-0.2.1/src/rh/cli.py +2778 -0
- rh_cli-0.2.1/src/rh/config.py +772 -0
- rh_cli-0.2.1/src/rh/doctor.py +143 -0
- rh_cli-0.2.1/src/rh/harnesses.json +37 -0
- rh_cli-0.2.1/src/rh/paths.py +178 -0
- rh_cli-0.2.1/src/rh/resume.py +119 -0
- {rh_cli-0.1.0 → rh_cli-0.2.1}/src/rh/search.py +11 -3
- rh_cli-0.2.1/src/rh/skills/research-helper/SKILL.md +667 -0
- rh_cli-0.2.1/src/rh/spawn.py +650 -0
- rh_cli-0.2.1/src/rh/tasks.py +309 -0
- {rh_cli-0.1.0 → rh_cli-0.2.1}/src/rh/ui.py +102 -3
- rh_cli-0.2.1/src/rh/workflow.py +408 -0
- rh_cli-0.2.1/src/rh_cli.egg-info/PKG-INFO +287 -0
- rh_cli-0.2.1/src/rh_cli.egg-info/SOURCES.txt +45 -0
- {rh_cli-0.1.0 → rh_cli-0.2.1}/src/rh_cli.egg-info/requires.txt +2 -0
- rh_cli-0.2.1/tests/test_cli.py +1675 -0
- rh_cli-0.2.1/tests/test_config_surface.py +135 -0
- rh_cli-0.1.0/PKG-INFO +0 -217
- rh_cli-0.1.0/README.md +0 -188
- rh_cli-0.1.0/src/rh/cli.py +0 -799
- rh_cli-0.1.0/src/rh/config.py +0 -308
- rh_cli-0.1.0/src/rh/paths.py +0 -96
- rh_cli-0.1.0/src/rh/resume.py +0 -186
- rh_cli-0.1.0/src/rh/skills/research-helper/SKILL.md +0 -497
- rh_cli-0.1.0/src/rh/tasks.py +0 -221
- rh_cli-0.1.0/src/rh/workflow.py +0 -268
- rh_cli-0.1.0/src/rh_cli.egg-info/PKG-INFO +0 -217
- rh_cli-0.1.0/src/rh_cli.egg-info/SOURCES.txt +0 -21
- rh_cli-0.1.0/tests/test_cli.py +0 -61
- {rh_cli-0.1.0 → rh_cli-0.2.1}/LICENSE +0 -0
- {rh_cli-0.1.0 → rh_cli-0.2.1}/setup.cfg +0 -0
- {rh_cli-0.1.0 → rh_cli-0.2.1}/src/rh_cli.egg-info/dependency_links.txt +0 -0
- {rh_cli-0.1.0 → rh_cli-0.2.1}/src/rh_cli.egg-info/entry_points.txt +0 -0
- {rh_cli-0.1.0 → rh_cli-0.2.1}/src/rh_cli.egg-info/top_level.txt +0 -0
|
@@ -0,0 +1,512 @@
|
|
|
1
|
+
# Changelog
|
|
2
|
+
|
|
3
|
+
All notable changes to `rh` (distribution `rh-cli`). Format follows
|
|
4
|
+
[Keep a Changelog](https://keepachangelog.com/1.1.0/); this project uses
|
|
5
|
+
[semantic versioning](https://semver.org/).
|
|
6
|
+
|
|
7
|
+
## [0.2.1] — 2026-11-19
|
|
8
|
+
|
|
9
|
+
Board edits, single-task inspection, and an upgrade path that keeps the checklist.
|
|
10
|
+
|
|
11
|
+
### Added
|
|
12
|
+
|
|
13
|
+
- **`rh tasks info <id>`** — one task or subtask in full: phase, owner, state,
|
|
14
|
+
deliverable, verify, risk, kind, timebox, dependencies, subtasks and timestamps
|
|
15
|
+
(`--json`).
|
|
16
|
+
- **`rh tasks rm <id> [--force]`** — drop a task (with its subtasks) or one subtask.
|
|
17
|
+
Refuses `working` work and tasks another task depends on unless `--force`;
|
|
18
|
+
surviving ids are never renumbered.
|
|
19
|
+
- **`rh tasks reorder <id> (--before|--after) <id>`** — move a task inside its phase;
|
|
20
|
+
list order is display order.
|
|
21
|
+
- **`.rh/ENVIRONMENT.md` records the `rh` version**, and the checklist's
|
|
22
|
+
`setup-rh-version` item flags a missing or stale record.
|
|
23
|
+
- **Upgrade procedure** (SKILL.md §10): the user upgrades the CLI, then
|
|
24
|
+
`rh init --force`, then `rh doctor`. It archives copies of `env.json` and
|
|
25
|
+
`tasks.json`, keeps notes/logs/tailored role files, and keeps checklist ticks.
|
|
26
|
+
|
|
27
|
+
### Changed
|
|
28
|
+
|
|
29
|
+
- **Hand-editing `.rh/tasks.json` is sanctioned for removal and reordering**
|
|
30
|
+
(SKILL.md §1): user approval first, valid JSON, list order is display order, ids
|
|
31
|
+
and `current_phase` untouched.
|
|
32
|
+
- **Board writes are strict.** An unparseable `tasks.json` makes `rh add task`,
|
|
33
|
+
`rh tasks set|phase|rm|reorder` fail with the file path and a `.rh/.bak/` hint
|
|
34
|
+
instead of silently starting from an empty board; reads still render, with a
|
|
35
|
+
warning naming the file.
|
|
36
|
+
- **`rh init` keeps manual checklist ticks** and recomputes generated ones; item
|
|
37
|
+
ids travel in hidden `<!-- id status -->` markers, and files written before this
|
|
38
|
+
release fall back to matching the item label.
|
|
39
|
+
- `rh info ENVIRONMENT.md` and the Orchestrator scaffold name the version record.
|
|
40
|
+
|
|
41
|
+
## [0.2.0] — 2026-11-18
|
|
42
|
+
|
|
43
|
+
A smaller command surface: one command per concept, panes that know their backend,
|
|
44
|
+
and a parser (typer) that makes the next change cheap. Several commands are gone —
|
|
45
|
+
see **Removed**.
|
|
46
|
+
|
|
47
|
+
### Added
|
|
48
|
+
|
|
49
|
+
- **`rh tasks phase [0-9]`** — the phase and its exit gate, and the gated setter,
|
|
50
|
+
now live where the work is. `--force` still records a waiver.
|
|
51
|
+
- **`rh tasks -c` / `--cancelled`** — cancelled tasks are hidden from the board by
|
|
52
|
+
default and counted in one footer line (`N tasks cancelled — check with rh tasks -c`).
|
|
53
|
+
- **`rh -t`** — short alias for `rh tasks` (next to the existing `rh -r`).
|
|
54
|
+
- **`rh agent list|info <role>|write [role]`** — one group for role files, with
|
|
55
|
+
`rh agent info` rendering the Markdown instead of dumping raw text, and
|
|
56
|
+
`rh agent status` unchanged.
|
|
57
|
+
- **Bounded `STATUS.md`.** History is the last ten one-line step transitions; older
|
|
58
|
+
lines move into that step's event log under `.rh/logs/`, and rule waivers get their
|
|
59
|
+
own `waivers:` list so trimming history never loses one. `.rh/CHECKLIST.md` flags a
|
|
60
|
+
longer history (`content-status-history`), and `rh info STATUS.md` states the cap.
|
|
61
|
+
- **Shipped pane templates per backend** — `agents.spawn` (yours, always wins) →
|
|
62
|
+
backend template (tmux, tuios, herdr, from `harnesses.json`) → detached process
|
|
63
|
+
with a log. `rh init` fills a blank `agents.spawn` from the chosen backend;
|
|
64
|
+
`rh doctor` reports a missing backend binary.
|
|
65
|
+
|
|
66
|
+
### Changed
|
|
67
|
+
|
|
68
|
+
- **BREAKING** The CLI parser is `typer` instead of `argparse`. Surviving commands,
|
|
69
|
+
flags and exit codes are unchanged except as listed below, and usage errors still
|
|
70
|
+
print the full help (with each command's one-line description) and exit 2.
|
|
71
|
+
- `--json` now means exactly one JSON document, per the CLI contract: `rh add task`
|
|
72
|
+
and `rh env list` print one document instead of prose/concatenated objects, and
|
|
73
|
+
`rh env set --json` sends its config warnings to stderr. `rh tasks --json`'s
|
|
74
|
+
`tasks` array follows the view (cancelled hidden unless `-c`); `summary` is
|
|
75
|
+
unchanged.
|
|
76
|
+
- `--root`/`--json` are accepted inside a command group too (`rh env --root X show`).
|
|
77
|
+
- **BREAKING** `-r`/`-t` expand only when the command line names no explicit
|
|
78
|
+
subcommand, so `rh -r status` (previously `rh resume` plus a flag it ignored) is
|
|
79
|
+
now a usage error.
|
|
80
|
+
- **BREAKING** `requires-python` is now `>=3.11` (was `>=3.9`); CI tests 3.11,
|
|
81
|
+
3.12 and 3.13.
|
|
82
|
+
|
|
83
|
+
### Removed
|
|
84
|
+
|
|
85
|
+
- **BREAKING** `rh phase` → `rh tasks phase` (same gates, same `--force`).
|
|
86
|
+
- **BREAKING** `rh gate check` → gate problems are printed by `rh tasks` and
|
|
87
|
+
`rh status`, and `rh tasks --json` carries the same `gate` object.
|
|
88
|
+
- **BREAKING** `rh agents [ROLE] [--write]` → `rh agent list` / `rh agent info <role>`
|
|
89
|
+
/ `rh agent write [role]`.
|
|
90
|
+
|
|
91
|
+
## [0.1.7] — 2026-11-12
|
|
92
|
+
|
|
93
|
+
Project-scoped agent names, a purpose question at intake, and a handoff that is
|
|
94
|
+
authored at session end instead of generated.
|
|
95
|
+
|
|
96
|
+
### Added
|
|
97
|
+
|
|
98
|
+
- **`{project}` placeholder and `rh-<project>-<role>` pane names.** Two projects
|
|
99
|
+
can run at the same time, so `agents.spawn` now names every pane/window with the
|
|
100
|
+
project slug: `tmux new-window -n rh-{project}-{agent} -c {root} {command}`
|
|
101
|
+
(e.g. `rh-optics-researcher`). The spawn prompt tells each agent the name it is
|
|
102
|
+
addressed by, and SKILL.md rule 15 makes it a rule. The old `rh-{agent}` template
|
|
103
|
+
updates in the README, `examples/` and `rh env list` help.
|
|
104
|
+
- **`## Purpose` in `PROBLEM.md`.** Phase 1 now starts by asking why this work and
|
|
105
|
+
what it is for (reproduce a reference, check data, a Q1 submission, a reproduction,
|
|
106
|
+
answering a reviewer …) — it decides what "done" means. `rh info PROBLEM.md`, the
|
|
107
|
+
checklist (`content-problem-purpose`) and the phase-1 gate (seven sections) agree.
|
|
108
|
+
- **`rh handoff` shows the next-session handoff; `rh handoff --clear` empties it.**
|
|
109
|
+
`.rh/HANDOFF.md` is created empty by `rh init`, written by the Orchestrator when a
|
|
110
|
+
session ends (what to do first, what to watch out for), read by `rh resume` only
|
|
111
|
+
when it has content, then cleared. `rh` no longer generates a handoff pack from
|
|
112
|
+
`STATUS.md`/`PLAN.md`/checklist — a handoff is a decision about the next move.
|
|
113
|
+
|
|
114
|
+
### Changed
|
|
115
|
+
|
|
116
|
+
- SKILL.md §5 gained a step-by-step spawn protocol (verify root, preview by name,
|
|
117
|
+
open, address by `rh-<project>-<role>`, close), plus the handoff template in §8.
|
|
118
|
+
The orchestrator role scaffold carries the same rules.
|
|
119
|
+
|
|
120
|
+
## [0.1.6] — 2026-11-10
|
|
121
|
+
|
|
122
|
+
A fifth task state for plans that die, role files that arrive populated, and the
|
|
123
|
+
PDF/arXiv tool named in the config. Plus fixes found by running the CLI against a
|
|
124
|
+
real project.
|
|
125
|
+
|
|
126
|
+
### Added
|
|
127
|
+
|
|
128
|
+
- **`literature.library` — the PDF library / arXiv download tool**, asked by the
|
|
129
|
+
`rh init` wizard (papr / zot / lit / none, with `✓ on PATH` per option) and settable
|
|
130
|
+
as `rh env set literature.library=papr`. `papr` is a TUI that keeps a paper library,
|
|
131
|
+
fetches from arXiv (`papr download <arxiv-id>`) and writes the manuscript; the skill
|
|
132
|
+
now says how it fits (point its `download_path` at `papers/`, library entry ≠ citation,
|
|
133
|
+
its Typst project may be the writing surface but `MANUSCRIPT.md` stays the record).
|
|
134
|
+
Wizard answers outside `agents.*`/`project.*` used to be dropped on write — they are
|
|
135
|
+
applied now, which is what lets this key be asked at init.
|
|
136
|
+
- **`cancelled` task state.** A later step can kill the plan, so a task can now be
|
|
137
|
+
dropped deliberately: `rh tasks set <id> cancelled` (also `--state cancelled` on
|
|
138
|
+
`rh add task`). It renders `⊘` and counts as *resolved* — the progress bars treat
|
|
139
|
+
it like complete — but the `complete` count never includes it, so the board still
|
|
140
|
+
says how much was actually delivered.
|
|
141
|
+
|
|
142
|
+
### Removed
|
|
143
|
+
|
|
144
|
+
- **`rh plan import` is gone**, with its parser. A plan is prose: people and agents
|
|
145
|
+
read it, nothing guesses at it. Tasks are created explicitly —
|
|
146
|
+
`rh add task "<title>" --id S0N --owner … --deliverable … --verify … --subtask "…"`
|
|
147
|
+
(repeatable `--subtask`) — by whoever owns the plan: the agent that writes it, or
|
|
148
|
+
the Orchestrator when the user writes it. `rh tasks set <id> working|complete`
|
|
149
|
+
moves them, subtask ids included. The bundled skill says the same at §1, §3
|
|
150
|
+
phase 0 and §10 contract 3, so nothing points at a command that no longer exists.
|
|
151
|
+
|
|
152
|
+
### Fixed
|
|
153
|
+
|
|
154
|
+
- **`rh init` wrote five blank role files.** `.rh/agents/` `literature`, `researcher`,
|
|
155
|
+
`reviewer`, `critique`, `writer` held `<what this agent is for — one sentence>`
|
|
156
|
+
fill-in blanks, so a fresh project had no role rules until someone typed them.
|
|
157
|
+
Every starter now carries its SKILL.md §5 row (owns / may / may not) plus the
|
|
158
|
+
session rules and the `{agent_file}` wiring note — deliberately generalized: the
|
|
159
|
+
job and its boundary, no project facts (paths, tools, phases, output names). A
|
|
160
|
+
`Tailor for this project` block in each file hands that narrowing to the
|
|
161
|
+
Orchestrator, who edits the file with the user. `rh agents --write` also
|
|
162
|
+
replaces a role file that still holds the old blank scaffold, and leaves an
|
|
163
|
+
edited one alone.
|
|
164
|
+
- `rh plan import` with no `PLAN.md` said "PLAN.md and the board agree", which is
|
|
165
|
+
wrong twice: nothing was compared and nothing was imported. It now says there is
|
|
166
|
+
no plan yet and points at `rh info PLAN.md` for the shapes it understands.
|
|
167
|
+
- `rh search` walked `.rh/.bak/` and `.rh/tmp/`, so a superseded file's line came
|
|
168
|
+
back ahead of the live one (`rh add` / `rh env set` archives made every query
|
|
169
|
+
noisier over time). Archives and transient render output are skipped now; logs
|
|
170
|
+
and notes are still searched — they are the project's written record.
|
|
171
|
+
|
|
172
|
+
## [0.1.4] — 2026-11-07
|
|
173
|
+
|
|
174
|
+
A plan written in ordinary Markdown, no planners if you don't want them, and the
|
|
175
|
+
project's software list in the config.
|
|
176
|
+
|
|
177
|
+
### Added
|
|
178
|
+
|
|
179
|
+
- **`PLAN.md` in general Markdown.** `rh plan import` used to accept one shape
|
|
180
|
+
(`### S07 — title`). It now recognises, first one found wins: a heading with an
|
|
181
|
+
explicit id at any level · any level>=3 heading · a numbered heading (`## 1. x`,
|
|
182
|
+
when no level>=3 heading exists) · a numbered line at column 0 · a bullet at
|
|
183
|
+
column 0. A heading that is not a step is a **section boundary**, so
|
|
184
|
+
`## 1. What works` never becomes a task. Steps without an id get positional ids
|
|
185
|
+
(`S01`…), fields are case-insensitive and all optional, and a plan with no
|
|
186
|
+
recognisable step now says so plus the accepted shapes instead of "the board
|
|
187
|
+
agrees".
|
|
188
|
+
- **Subtasks are read from `PLAN.md`.** Bullet lines inside a step (`- [ ] x`,
|
|
189
|
+
indented or not) become that step's subtasks. Import creates them; re-import
|
|
190
|
+
**appends** new ones (`S07.3`) and never deletes one or resets a state — a
|
|
191
|
+
subtask that leaves the plan keeps its row and its evidence.
|
|
192
|
+
- **`agents.planners.models = none`.** `none` / `0` / `[]` means no planner is
|
|
193
|
+
configured and **the user is the planner**: phase 4's exit gate reads
|
|
194
|
+
"PLAN.md: written by the user (no planner models configured) + user approval",
|
|
195
|
+
`rh doctor` and `rh env show` say so, and `rh info PLAN.md` names the user as
|
|
196
|
+
author with the Orchestrator reconciling. Before, the key was not settable at
|
|
197
|
+
all (`rh env set` knew only `.0` / `.1`).
|
|
198
|
+
- **`project.environment`** — the software this project actually runs, as a JSON
|
|
199
|
+
list of `{name, version, note}`. `rh env set` writes it, `validate()` names an
|
|
200
|
+
entry with no `name`, and `rh env show` prints a *Project environment* table.
|
|
201
|
+
List coercion passes object entries through untouched instead of stringifying
|
|
202
|
+
them.
|
|
203
|
+
- `rh plan import` reports how many subtasks a new step carries.
|
|
204
|
+
|
|
205
|
+
### Changed
|
|
206
|
+
|
|
207
|
+
- Phase-exit text moved behind `tasks.phase_exit(phase, user_plans)`; `rh phase`,
|
|
208
|
+
its gate-refusal messages and the checklist labels all use it, so no surface
|
|
209
|
+
can disagree about who writes `PLAN.md`.
|
|
210
|
+
- `rh info PLAN.md` documents the accepted shapes, the subtask rule, and who is
|
|
211
|
+
responsible for keeping the board in sync (an agent that writes the plan
|
|
212
|
+
imports its own steps; a user-written plan is imported by the Orchestrator).
|
|
213
|
+
- `rh show` no longer half-strips backticks from a step title — inner markup is
|
|
214
|
+
the author's.
|
|
215
|
+
|
|
216
|
+
### Fixed
|
|
217
|
+
|
|
218
|
+
- A plan parsed as bullets-only when its steps were numbered `## 1.` headings
|
|
219
|
+
(the bullets became steps). Covered by a test that fails on 0.1.3.
|
|
220
|
+
- `agents.planners.models` holding a non-model entry was silently accepted.
|
|
221
|
+
|
|
222
|
+
Tests: 83 passed (63 before this change; 20 new in `tests/test_plan_lenient.py`).
|
|
223
|
+
|
|
224
|
+
## [0.1.3] — 2026-11-07
|
|
225
|
+
|
|
226
|
+
Fixes from the second test pass: one config file, a real Orchestrator to-do list
|
|
227
|
+
at init, role files at init, and a silent `rh resume`.
|
|
228
|
+
|
|
229
|
+
### Changed (breaking)
|
|
230
|
+
|
|
231
|
+
- **`.rh/ENV.md` is gone.** `.rh/env.json` is the single config file: `rh env set`
|
|
232
|
+
writes it, `rh env show` reads it, and nothing wraps JSON in Markdown anymore.
|
|
233
|
+
The Orchestrator's free-form environment notes moved to the phase-2 artifact
|
|
234
|
+
`.rh/ENVIRONMENT.md`, so config and prose never share a file. A legacy
|
|
235
|
+
`.rh/ENV.md` (or a root `ENV.md`) is folded into `env.json` +
|
|
236
|
+
`ENVIRONMENT.md` on first read and the Markdown is archived to
|
|
237
|
+
`.rh/.bak/<timestamp>/` — never deleted.
|
|
238
|
+
- `rh env show` prints the config plus `.rh/ENVIRONMENT.md`; `rh doctor` checks
|
|
239
|
+
`env.json` (`env-json` instead of `env-md`).
|
|
240
|
+
- Phase gates, `rh info`, `rh search`, `rh handoff` and `rh workflow` name
|
|
241
|
+
`.rh/ENVIRONMENT.md` / `.rh/env.json`.
|
|
242
|
+
- **`rh spawn-terminal` is gone.** `rh spawn terminal <cmd>` is the only form
|
|
243
|
+
(the extra alias is no longer registered or listed).
|
|
244
|
+
- **`rh checklist` is gone.** The generated `.rh/CHECKLIST.md` is the checklist;
|
|
245
|
+
the orchestrator role file (its additional system prompt) points the agent at
|
|
246
|
+
it, so there is no separate command to remember. `rh init` still writes the
|
|
247
|
+
file, and `rh handoff` still packs it.
|
|
248
|
+
|
|
249
|
+
### Added
|
|
250
|
+
|
|
251
|
+
- **`rh init` creates the Orchestrator's checklist** at `.rh/CHECKLIST.md`: setup
|
|
252
|
+
tasks (missing environment values to ask about, launch commands to wire, role
|
|
253
|
+
files to tailor, `workflow.mmd` after the plan) followed by every phase exit
|
|
254
|
+
gate and the SKILL.md §0 rules, each checked against disk. The orchestrator
|
|
255
|
+
role file tells the agent to read it first, so no CLI command is involved.
|
|
256
|
+
- **`rh init` creates the default role files** (`.rh/agents/orchestrator.md` +
|
|
257
|
+
the starter roles) instead of leaving `rh agents --write` as a hidden step.
|
|
258
|
+
The orchestrator file carries the actual contract (owns/may/may-not, the
|
|
259
|
+
checklist, rules 9/12/13), so a fresh session knows its job immediately.
|
|
260
|
+
- **A rich `rh init` questionnaire**: key hints once, then one compact question
|
|
261
|
+
frame redrawn in place (no full-screen clear). `rich.prompt` validates the
|
|
262
|
+
answer, numbered options list the agent CLIs, shells and pane backends that
|
|
263
|
+
are actually on PATH (`other` types one that is not), and Enter keeps the
|
|
264
|
+
default, Tab submits and moves on (readline/libedit binding), `s` skips the
|
|
265
|
+
key, `q` finishes with the answers so far, Ctrl-C skips the rest. Skills are
|
|
266
|
+
not asked here — that is the Orchestrator's job (`rh env set agents.skills=`).
|
|
267
|
+
- **Harness start templates** (`src/rh/harnesses.json`, data only): the init
|
|
268
|
+
questionnaire's harness answer fills `agents.cli` and
|
|
269
|
+
`agents.commands.orchestrator|default` where still blank, so `rh resume` works
|
|
270
|
+
from a fresh init. Existing commands are never overwritten.
|
|
271
|
+
|
|
272
|
+
### Fixed
|
|
273
|
+
|
|
274
|
+
- **The `pi` harness template used flags `pi` does not have.** `--prompt-file`
|
|
275
|
+
does not exist, and `--prompt-template` is not a prompt either — it registers a
|
|
276
|
+
`/command` and sends no message, so `rh resume` opened `pi` with the prompt
|
|
277
|
+
only loaded as a slash command. The launch template now passes the prompt
|
|
278
|
+
**positionally** (`{prompt}`), and the role file goes in through
|
|
279
|
+
`--append-system-prompt {agent_file}` — never `--system-prompt`, which would
|
|
280
|
+
replace `pi`'s own system prompt. Same fix in `examples/envs/pi.sh`,
|
|
281
|
+
`examples/configure-env.sh`, README and the bundled SKILL.md.
|
|
282
|
+
- **The orchestrator is no longer handed a skill restriction.** Its launch
|
|
283
|
+
template carries no `--skill`: the Orchestrator must see every skill to decide
|
|
284
|
+
which one each spawned agent is assigned. `rh` also stopped warning that
|
|
285
|
+
"skills are not restricted at the CLI" for the orchestrator role.
|
|
286
|
+
- **Bad CLI usage falls back to the full `--help`** and nothing else — no
|
|
287
|
+
`error: invalid choice …` line (exit 2). `rh --help` and `rh --version` exit 0.
|
|
288
|
+
- **`rh resume` no longer prints the prompt** before opening the harness. Its
|
|
289
|
+
first message is the skill read (“Load the `research-helper` skill …”), and it
|
|
290
|
+
opens the harness with `.rh/agents/orchestrator.md` wired in through
|
|
291
|
+
`{agent_file}`. If a hand-written template omits `{agent_file}`, the role file
|
|
292
|
+
is prepended to the prompt instead, so the Orchestrator knows its job from the
|
|
293
|
+
get go.
|
|
294
|
+
|
|
295
|
+
## [0.1.2] — 2026-11-07
|
|
296
|
+
|
|
297
|
+
Delegation, config honesty and artifact documentation (from the ISSUES list).
|
|
298
|
+
|
|
299
|
+
### Added
|
|
300
|
+
|
|
301
|
+
- **`rh spawn <role> [prompt]`** — open one role agent (`literature`,
|
|
302
|
+
`researcher`, `reviewer`, `critique`, `writer`, or any role you define) with
|
|
303
|
+
the launch command written in `.rh/ENV.md`. It passes the role file
|
|
304
|
+
`.rh/agents/<role>.md` and appends the session facts (model, skills, its board
|
|
305
|
+
tasks with deliverable + verify, the current phase) and opens it through `agents.spawn` (tmux/herdr/tuios pane) or as a
|
|
306
|
+
detached process logging to `.rh/logs/`. Flags: `--prompt`, `--prompt-file`,
|
|
307
|
+
`--inline`, `--dry-run`, `--json`.
|
|
308
|
+
- **`rh info [FILE.md]`** — one clean description per artifact: purpose, owner,
|
|
309
|
+
phase, required sections, template and the rules that bite there. Accepts
|
|
310
|
+
`status`, `STATUS`, `.rh/STATUS.md`; `--template` prints just the template,
|
|
311
|
+
`--json` for agents, no argument lists every artifact.
|
|
312
|
+
- **`agents.commands.<role>`** and **`agents.spawn`** config keys: `rh` hardcodes
|
|
313
|
+
no agent flag. Templates take `{model}` `{thinking}` `{skills}` `{prompt}`
|
|
314
|
+
`{prompt_file}` `{root}` `{slug}` `{role}` (and `{command}` in
|
|
315
|
+
`agents.spawn`); a placeholder with an empty value drops the flag that
|
|
316
|
+
introduced it. Any role name is settable: `rh env set
|
|
317
|
+
agents.commands.<role>=…`, `rh env set agents.roles.<role>.model=…`.
|
|
318
|
+
- **`thinking`** level per role: `agents.roles.<role>` now takes
|
|
319
|
+
`{provider, model, thinking}`, shown as `pi/glm-5 (thinking: high)`.
|
|
320
|
+
- `rh resume --prompt-file FILE` for this session's instructions.
|
|
321
|
+
- `examples/` — starter scripts (`configure-env.sh`, `spawn-researcher.sh`,
|
|
322
|
+
`spawn-critique.sh`) and per-role prompt templates with `{{PLACEHOLDER}}`s, so
|
|
323
|
+
an agent (or a user) can see how commands, roles, tasks and prompts fit.
|
|
324
|
+
|
|
325
|
+
- **`rh spawn terminal <command>`** — open a plain shell terminal (not an agent) in a
|
|
326
|
+
new pane, e.g. `rh spawn terminal 'pytest -q'`. Shell from `agents.shell` in
|
|
327
|
+
`.rh/ENV.md` (else `$SHELL`, else `sh`); pane layout from `agents.split`
|
|
328
|
+
(`vertical` | `horizontal` | `window` — new workspace/window), asked of the user
|
|
329
|
+
when unset. `--dry-run`, `--split`, `--json`.
|
|
330
|
+
- **Per-role agent files** — `.rh/agents/<role>.md` is the *only* role
|
|
331
|
+
description in the system: no role info is hardcoded in `rh`. `rh agents
|
|
332
|
+
[ROLE] [--write]` lists/shows/scaffolds them. The Orchestrator edits them
|
|
333
|
+
with user approval and records the launch command per role in `.rh/ENV.md`
|
|
334
|
+
(`agents.commands.<role>`), which `rh spawn` uses exactly; the `{agent_file}`
|
|
335
|
+
placeholder puts the role file into the harness's system-prompt area
|
|
336
|
+
(e.g. `--system-prompt-file {agent_file}` or the harness equivalent).
|
|
337
|
+
- **`rh checklist`** — the Orchestrator's live checklist: every SKILL.md §0
|
|
338
|
+
non-negotiable plus each phase's exit gate, checked against the project on
|
|
339
|
+
disk (via `rh phase` gates), never self-reported. `--json` for agents.
|
|
340
|
+
v2 adds content checks (STATUS header, PLAN `verify:` lines, DOI per
|
|
341
|
+
literature entry, exactly one layout).
|
|
342
|
+
- **`rh doctor`** — verifies what ENV.md names: shell on PATH, `agents.split`
|
|
343
|
+
valid, every `agents.commands.<role>` splittable with `{agent_file}` +
|
|
344
|
+
`{prompt}`, role files present, backend binary present. Checks by name only.
|
|
345
|
+
- **`rh agent status`, `rh gate check`, `rh log tail`, `rh handoff`** — spawn
|
|
346
|
+
records in `.rh/logs/spawns.jsonl`, machine-readable gate report, §7 log
|
|
347
|
+
reader, fresh-session pack (rule 14).
|
|
348
|
+
- **`agents.count` budget enforcement** — `rh spawn` refuses the (count+1)th
|
|
349
|
+
concurrent agent without `--force` (waiver → STATUS.md, rule 11).
|
|
350
|
+
- **List/content placeholders** — `{skills:list}` expands to one token per
|
|
351
|
+
skill; `{agent_file_contents}` puts the role file contents inline.
|
|
352
|
+
- **`{agent_file}` at `rh env set` time** — missing `{agent_file}` or
|
|
353
|
+
`{prompt}` in a command template is a config warning immediately, not only
|
|
354
|
+
at spawn.
|
|
355
|
+
- **`examples/envs/{pi,claude,codex}.sh`** — the only place harness flags live.
|
|
356
|
+
- **`rh init` takes no project name.** It initializes the directory you run it
|
|
357
|
+
in and names the project after that directory (a label only — no directory is
|
|
358
|
+
created, no `--slug` flag). Rename it with `rh env set project.name=...`; the
|
|
359
|
+
questionnaire may override it. `project.slug` is read as `project.name` and
|
|
360
|
+
migrated on the next write.
|
|
361
|
+
- **`rh init` is non-destructive.** Existing `ENV.md`/`tasks.json` block init
|
|
362
|
+
unless `--force`, which archives a *copy* to `.rh/.bak/<timestamp>/` first
|
|
363
|
+
(originals and their environment notes stay put). `--clean` archives the whole
|
|
364
|
+
`.rh/` tree — event logs and notes included — before recreating it, so nothing
|
|
365
|
+
is deleted that was not archived. Backups never collide inside one second.
|
|
366
|
+
`--dry-run` writes nothing. `PLAN.md` is never touched by init.
|
|
367
|
+
- **`rh init` questionnaire.** Few skippable questions (slug, count, backend,
|
|
368
|
+
CLI, shell, split, skills, harness snippet). Defaults are read from the
|
|
369
|
+
ENV.md already on disk, so Enter keeps the project's settings instead of
|
|
370
|
+
resetting them; `s` really skips (the key is not written) and `Ctrl-C` ends
|
|
371
|
+
the questionnaire; skipped questions become the Orchestrator's job. The
|
|
372
|
+
overwrite guard runs *before* the questions, so a configured project is not
|
|
373
|
+
quizzed and then refused. `--non-interactive` / `--json` skip it entirely.
|
|
374
|
+
- **`PLAN.md` guard.** Pre-existing `PLAN.md` is adopted, never overwritten:
|
|
375
|
+
`rh` reads it and mirrors it with `rh plan import`; forced writes archive
|
|
376
|
+
to `.rh/.bak/` first. `rh plan import` stays read-only for `PLAN.md`.
|
|
377
|
+
|
|
378
|
+
### Changed
|
|
379
|
+
|
|
380
|
+
- **`rh resume` now starts the session `.rh/ENV.md` describes**: command from
|
|
381
|
+
`agents.commands.orchestrator`, model + thinking level from
|
|
382
|
+
`agents.roles.orchestrator`, and only the `agents.skills` list loaded (fewer
|
|
383
|
+
skills, smaller context). It prints the model, skills and the exact command
|
|
384
|
+
before exec'ing, and `--dry-run`/`--json` return them without spawning. When
|
|
385
|
+
the command template omits `{model}`, `{thinking}` or `{skills}`, `rh` warns
|
|
386
|
+
that the ENV.md settings are not being honoured instead of guessing a flag.
|
|
387
|
+
- `rh resume`/`rh spawn` prompts list the allowed skills explicitly and point at
|
|
388
|
+
`rh info <FILE>` and `rh spawn` for delegation.
|
|
389
|
+
- The `agents.cli` fallback survives only for projects with no command template:
|
|
390
|
+
`rh resume` asks for a binary, persists it and warns that model/thinking/skills
|
|
391
|
+
are not passed.
|
|
392
|
+
|
|
393
|
+
### Fixed
|
|
394
|
+
|
|
395
|
+
- **`DEFAULT_ENV` was mutated by `config.load_env()`.** `deep_merge` shared nested
|
|
396
|
+
dicts with the defaults, so any `env set` on a key missing from ENV.md edited
|
|
397
|
+
the process-wide defaults for everything that ran afterwards. Now deep-copied.
|
|
398
|
+
- **`rh init` skip wrote `"s"`.** `s` was stored as the value: `project.name="s"`,
|
|
399
|
+
`agents.backend="s"`, `skills=["s"]` — an invalid config. Skip is now absent.
|
|
400
|
+
- **`rh init --clean` deleted without archiving** (and `--clean --force` archived
|
|
401
|
+
nothing, since the removal ran first). The archive now happens before any
|
|
402
|
+
deletion, covers the whole `.rh/` tree, and keeps the archives of earlier
|
|
403
|
+
cleans.
|
|
404
|
+
- **`rh checklist` header check** missed a bold header (`phase: **6 IMPLEMENT**`)
|
|
405
|
+
and claimed to verify `step:`/`updated:` while only testing `phase:`; it now
|
|
406
|
+
names the missing part. Phases after the current one report `later` instead of
|
|
407
|
+
counting as pending.
|
|
408
|
+
- **`rh doctor`** now also checks the role file for every role that has a model
|
|
409
|
+
configured, not only roles with a command template.
|
|
410
|
+
- **`--dry-run` wrote prompt files** (`spawn-*.md`, `resume-prompt.md`). It now
|
|
411
|
+
writes nothing.
|
|
412
|
+
|
|
413
|
+
- **Duplicate progress bars.** `rh status` printed the same completion twice (a
|
|
414
|
+
`Progress:` line plus the bar under the phase rail) and repeated the counts in
|
|
415
|
+
an openspec-style lane block, where idle tasks also rendered as `[x]`. Now:
|
|
416
|
+
one rail, one progress line, then the per-phase breakdown.
|
|
417
|
+
- `rh tasks` printed a second bar for subtasks; the subtask count is now a line
|
|
418
|
+
and each task row already carries `[done/total subtasks]`.
|
|
419
|
+
- `rh describe` reports the artifacts, `rh info`, and the spawn commands/
|
|
420
|
+
placeholders so an agent can find them without reading the source.
|
|
421
|
+
|
|
422
|
+
## [0.1.1] — 2026-11-04
|
|
423
|
+
|
|
424
|
+
Config, orchestration and task-board cleanup driven by real-project use
|
|
425
|
+
(`jmcc_cepo4_gcn`).
|
|
426
|
+
|
|
427
|
+
- **`agents.skills`** config key (list): the skills every spawned agent must read
|
|
428
|
+
before touching the project. Defaults to `["research-helper"]`, settable with
|
|
429
|
+
`rh env set agents.skills=research-helper,crw`.
|
|
430
|
+
- **Built-in ASCII workflow view.** When no external ASCII renderer is installed,
|
|
431
|
+
`rh workflow` renders the agent-authored `.rh/workflow.mmd` itself: nodes
|
|
432
|
+
grouped by subgraph with state glyphs (done / current / gate / conditional /
|
|
433
|
+
todo) plus the edge list, instead of ending in "no renderer".
|
|
434
|
+
- Renderer **priority list** shared by `rh workflow`: `bm` (beautiful-mermaid
|
|
435
|
+
CLI, also found as `beautiful-mermaid`) first, then `mermaid-ascii`,
|
|
436
|
+
`mermansi`, `mermaid2term`, `merx`, `ma`, then the built-in view. For
|
|
437
|
+
`--svg`/`--png`: `bm`, `mmdr`, `mmdc`.
|
|
438
|
+
- `rh tasks set <id> <state>` (task **and** subtask ids); `rh add task --subtask`
|
|
439
|
+
to record subtasks at creation time.
|
|
440
|
+
- `rh tasks` now prints every subtask with its own state, not just a count.
|
|
441
|
+
- `rh phase` with no argument shows the current phase, its exit gate and whether
|
|
442
|
+
it is met, plus the next phase.
|
|
443
|
+
|
|
444
|
+
### Changed
|
|
445
|
+
|
|
446
|
+
- **`ENV.md` now lives at `.rh/ENV.md`**, never at the project root. A root
|
|
447
|
+
`ENV.md` or a `.rh/ENV.md` symlink pointing at one is migrated automatically
|
|
448
|
+
on first read/write.
|
|
449
|
+
- **Downloaded papers stay in `./papers/`.** The CLI no longer creates or reads
|
|
450
|
+
`.rh/papers/` and never deletes a paper.
|
|
451
|
+
- **`rh resume` runs the agent in the terminal the command was called from**
|
|
452
|
+
(`exec` in place, after printing the project path) instead of opening a new
|
|
453
|
+
tmux window / herdr tab / tuios window. Spawning extra agents is the
|
|
454
|
+
Orchestrator's job via `agents.backend`.
|
|
455
|
+
- `rh resume` refuses to start unless the resolved root has `.rh/` and a
|
|
456
|
+
parseable `.rh/ENV.md`, and prints the path it is about to use.
|
|
457
|
+
- `rh status` reports progress only: per-phase bars with the working ids. Agent
|
|
458
|
+
configuration is no longer printed (or emitted in `--json`), and task/subtask
|
|
459
|
+
detail points to `rh tasks`.
|
|
460
|
+
- `rh workflow --tools` renders a proper table (renderer, state, binary aliases,
|
|
461
|
+
install hint) instead of one unreadable line.
|
|
462
|
+
- `rh plan import` treats `PLAN.md` as truth: existing ids get their title,
|
|
463
|
+
owner, deliverable, verify, kind, risk, timebox and dependencies refreshed
|
|
464
|
+
(state and subtasks are kept), and it reports `+ added` / `~ updated` with the
|
|
465
|
+
changed fields.
|
|
466
|
+
- Malformed `phase` values in `.rh/tasks.json` (e.g. `"6 IMPLEMENT"`) are
|
|
467
|
+
normalised on load, so mixed task data no longer crashes `rh status`/`rh tasks`.
|
|
468
|
+
- Task dependency parsing ignores the `none` placeholder.
|
|
469
|
+
- `rh init` output reflects the new layout (`.rh/ENV.md`, root `papers/`, no
|
|
470
|
+
`.rh/papers/`).
|
|
471
|
+
|
|
472
|
+
### Fixed
|
|
473
|
+
|
|
474
|
+
- `rh workflow` did not use an installed ASCII renderer at all when its output
|
|
475
|
+
could be wide: the early "no tools" return skipped the fallback and the `bm`
|
|
476
|
+
invocation used a non-existent `--padding-y` flag. Both fixed; output wider
|
|
477
|
+
than the terminal falls back to the built-in view with the reason shown.
|
|
478
|
+
- `rh plan import` did not recognise suffixed step ids such as **`S04b`**, and
|
|
479
|
+
never refreshed titles, so retired titles stayed on the board.
|
|
480
|
+
- `rh plan import` lost newly added steps instead of saving them.
|
|
481
|
+
- `rh task …` is removed; `rh tasks` is the single board command.
|
|
482
|
+
|
|
483
|
+
### Docs
|
|
484
|
+
|
|
485
|
+
- `SKILL.md` rewritten around the Orchestrator: it is always the current runner;
|
|
486
|
+
ask the user how many **parallel agents** they want for planning and research;
|
|
487
|
+
the two planners argue with each other and hand a finalized plan to the
|
|
488
|
+
Orchestrator; the single-agent case hands the user to a fresh session at
|
|
489
|
+
gate-heavy steps; spawned agents are closed when their job is done; the project
|
|
490
|
+
path is verified before any agent is opened; `STATUS.md` is updated by the
|
|
491
|
+
Orchestrator itself; `.rh/ENV.md` and `./papers/` locations updated.
|
|
492
|
+
- `README.md` / `ROADMAP.md` updated for the command and layout changes.
|
|
493
|
+
|
|
494
|
+
## [0.1.0] — 2026-11-04
|
|
495
|
+
|
|
496
|
+
First published release (`rh-cli` on PyPI; the earlier PyPI name was confusable
|
|
497
|
+
with `researchhelper`).
|
|
498
|
+
|
|
499
|
+
- `.rh/` workspace, config block as the single config source + `.rh/env.json`
|
|
500
|
+
mirror, `rh env show|set|list`, `rh describe`, `rh init`.
|
|
501
|
+
- Heading-structured artifacts (`rh add note|literature|layout|gaps|problem|
|
|
502
|
+
plan|status|manuscript|log|task`) and rg-first `rh search`.
|
|
503
|
+
- Gated phase pipeline (`rh phase`) with waivers, `rh status`, `rh tasks`,
|
|
504
|
+
`rh plan import`.
|
|
505
|
+
- `rh workflow` viewer for the agent-authored `.rh/workflow.mmd`.
|
|
506
|
+
- `rh resume` / `rh -r`, `rh skill install`, bundled `research-helper` skill.
|
|
507
|
+
- rich-based UI with `--json` on every command.
|
|
508
|
+
|
|
509
|
+
[0.1.3]: https://github.com/hitaishi2222/research-helper/compare/v0.1.2...v0.1.3
|
|
510
|
+
[0.1.2]: https://github.com/hitaishi2222/research-helper/compare/v0.1.1...v0.1.2
|
|
511
|
+
[0.1.1]: https://github.com/hitaishi2222/research-helper/compare/v0.1.0...v0.1.1
|
|
512
|
+
[0.1.0]: https://github.com/hitaishi2222/research-helper/releases/tag/v0.1.0
|
rh_cli-0.2.1/MANIFEST.in
ADDED
|
@@ -0,0 +1,7 @@
|
|
|
1
|
+
# sdist extras — the docs and the agent-facing examples ship with the source
|
|
2
|
+
# distribution so `pip download rh-cli --no-binary :all:` gets them too.
|
|
3
|
+
include README.md
|
|
4
|
+
include CHANGELOG.md
|
|
5
|
+
include ROADMAP.md
|
|
6
|
+
include LICENSE
|
|
7
|
+
recursive-include examples *.md *.sh
|