SkillTide Submit
Official descriptionOpen source7d ago
DraconDev/pi-goal-list-loop-audit

DraconDev/pi-goal-list-loop-audit

Long-running autonomy for Pi: draft goals, build projects from specs, run audited task queues, and recover interrupted work. Independent auditors verify completion.
View on GitHub ↗Direct link · no redirect page
Star count
27 ★ · DraconDev/pi-goal-list-loop-audit · read from the repository page · (checked 2026-10-16 20:00:20)
Source
read directly from its GitHub page · type: open-source project
Harvest history
first seen 2026-10-10 · last seen 2026-10-14 · seen 2× · listed by 1 sources
Score
score 103 · 27 ★ · 1 sources · repo
FAQ (answers come from our records)
How do I install it?
We did not find an install command for this repository (our install coverage is only about 4%). Check its README or releases page.
Who is it for?
We filed it under these tags: 开源项目. The tags come from the entry itself, not a taxonomy we imposed.
Who maintains it?
repository DraconDev/pi-goal-list-loop-audit;first catalogued 2026-10-10;last seen 2026-10-14。There is no official certification or quality rating here — we record sources and dates only.
Third-party quality signals

pi-goal-list-loop-audit

以下摘录自该仓库的 README(片段) · 在 GitHub 读完整 README →

> **Give Pi an outcome—not another prompt to stop halfway through.**
GLLA turns substantial work into a durable, supervised process: agree on the
finish line, keep working across turns, recover interruptions, and check the
result before calling it complete.
Then, in Pi:
Review the proposed objective and acceptance criteria, confirm them, and let
Pi work. Use /glla status to see what it is doing and what needs your attention.
**Waiting** yields to named background work while supervision remains active;
**paused** freezes automatic continuation. See waiting and recovery.
Install and update · Practical workflows ·
Recovery · Settings ·
Pi package catalog
About GLLA
Coding agents are good at taking the next step. Substantial projects need
something more: a remembered commitment, a reason to keep going, a way to
recover, and a finish line that does not depend on the agent sounding confident.
GLLA supplies that supervision inside Pi.
The main agent researches and implements; GLLA keeps the objective, progress,
pauses and completion claims on disk. A separate auditor inspects the result
against the agreed contract. Rejected claims return to work rather than becoming
successful-looking endings.
This is useful for features spanning several subsystems, migrations, repository
audits, documentation overhauls, backlogs and projects built from a specification.
It is not a new model, a hosted coding service, or a promise of hands-off correctness.
It works with your Pi session, tools and authenticated providers.
**The ambition is less babysitting, not less accountability.** You choose the
outcome and permissions; the agent handles ordinary implementation decisions;
important scope changes and decisions remain visible. Durable state and evidence
make it easier to inspect what happened—even after a long run or interruption.
Pick the right work shape
| You want… | Use | What counts as finished |
|---|---|---|
| One result: a fix, feature, migration, research report or audit | /goal | An accepted completion audit against the objective and contract |
| Several independently verifiable results | /list | Each item is audited separately; the queue advances after settlement |
| Repeated improvement | /loop | A configured bound, metric plateau or explicit stop—not a claim that everything is solved |
| A project built toward an evolving specification | /loop respec | Every adopted requirement independently verified and the project archived |
For a quick question or tiny edit, ordinary Pi chat is usually enough. Use GLLA
when continuity and verification are worth the extra model calls and coordination.
Start with one goal
A rough request can become a confirmed plan:
GLLA drafts the outcome, asks questions when needed and offers confirmation.
Bare /goal opens the interview; /goal plan ... requests research-first drafting.
You do not need to arrive with a perfect specification.
If you already know the finish line, a Done when: contract starts directly:
/goal start ... explicitly skips drafting. Be deliberate: an explicit start
can replace existing work. Drafting and direct starts are different consent paths.
The work cycle
1. **Agree:** preserve the objective and observable acceptance criteria.
2. **Build:** the agent researches, plans and implements across supervised turns.
3. **Recover:** retain work and use bounded recovery when providers, sessions or workers fail.
4. **Claim:** the agent submits evidence; saying “done” in chat does not close the goal.
5. **Audit:** a detached verifier checks the saved claim. Missing evidence or rejection keeps work open.
6. **Archive:** accepted, durably settled work leaves a result and evidence record.
The agent can propose a task plan with propose_task_list; confirming the plan
makes its tasks tracked milestones, not proof of correctness. Completion claims
with unfinished tasks are refused unless an explicit deferral was recorded.
Controls you will actually use
| Command | Purpose |
|---|---|
| /glla status | Inspect supervision, recovery and current work |
| /goal status | Inspect the saved goal |
| /glla pause | Freeze automatic supervision without killing an already running tool |
Source
github · https://github.com/DraconDev/pi-goal-list-loop-audit
Browse by field
Open source
Traction
27 ★
added
3d ago

More in Open source

All 731 in Open source →

Filed under

Every tag opens the full list for that filter. Counts are the live catalogue figures.

Browse all fields →

← Back to index