agent-build-propose
Skill1 starsChanged 5 months agoApache-2.0View on GitHub
Create or revise a review-ready Agentic Build proposal.md for one Truffle change packet. Use when explicitly asked to run agent-build-propose or draft an Agentic Build proposal; stop before implementation.
In this repository10 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- agent-build-decideSkill
- agent-build-reviewSkill
- agent-learning-mineSkill
- commit-allSkill
- learning-reviewSkill
- market-read-labSkill
- review-changeSkill
- work-menuSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
13 rules. About 0 are common, 4 are distinctive, and 9 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
9
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 12
Create or revise one review-ready, plan-only proposal.md for a bounded Agentic Build change packet.
- Distinctive: line 18
Before writing, do your work "offstage".
- Very specific: line 21
Check whether a packet is warranted: If the request is an obvious low-risk direct fix, already answered by prior work, or too unclear to …
- Distinctive: line 31
Read these before writing:
- Distinctive: line 38
Read additional docs when the proposed change touches their authority surface.
- Very specific: line 42
If the user gives an existing packet path, update that packet's proposal.md.
- Very specific: line 45
Use proposal-template.md as the structure.
- Very specific: line 49
Only create or update the packet proposal.md.
- Very specific: line 51
Keep frame context inside proposal.md.
- Very specific: line 53
If a required field is unknown, write unknown plus why.
- Distinctive: line 55
Prefer the simplest option that can meet the acceptance checks this proposal defines.
- Very specific: line 56
Follow the size defaults in 2026-06-21-lead-context.md.
- Very specific: line 57
Treat high risk as planning-only unless Brian explicitly approves implementation later.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
$ARGUMENTS
Purpose
Create or revise one review-ready, plan-only proposal.md for a bounded Agentic Build change packet.
Agentic Build is not the default for every task. Its job is to preserve reasoning, compare approaches, gate risky changes, or keep a cut/park record when that weight is earned.
Work Protocol
Before writing, do your work "offstage". Do not narrate this checklist unless it changes the recommendation.
- Frame the capability, not just the request. Identify the real job, success condition, non-goal, and what decision this proposal should unlock. If the request could mean materially different jobs, ask one clarifying question before writing.
- Check whether a packet is warranted. If the request is an obvious low-risk direct fix, already answered by prior work, or too unclear to frame inside
proposal.md, stop and report the smaller next move instead of manufacturing a proposal. - Stretch before you shrink. In scratch, consider several solution families before converging: the boring/small fix; the best prior-art path; a new-source or new-tool path; and a more ambitious version that would change the system's capability if it worked. Prior work is evidence, not the ceiling.
- Find the smallest meaningful slice. For each promising family, ask: what is the smallest probe, MVP, or reversible change that would move us forward and teach something real? Prefer the slice that tests the important uncertainty, not the slice that merely creates the least surface area.
- Check prior art. Search targeted local surfaces before proposing new machinery:
experiments/,_design/,BACKLOG.md, relevantskills/, relevanttools/, and existing packets. Ask what each prior artifact proves, what it falsifies, and what it did not explore. Mention prior art only when it affects the recommendation, risk, or scope. - Name the load-bearing value claim. What must be true for this proposal to be worth building? If the proposal uses new tools, sources, or mechanisms, name the bet they are making.
- Define the cheapest disconfirming check. What small test, fixture, dry run, known case, or review would reveal that the proposal is well-formed but useless?
- Converge hard. Recommend the simplest path that preserves the upside and can pass the value check. Compare only the strongest 2-3 finalist options in
proposal.md; keep discarded scratch options offstage unless one changes the recommendation.
Required Context
Read these before writing:
experiments/01-agentic-build/README.mdexperiments/01-agentic-build/2026-06-21-proposal.mdexperiments/01-agentic-build/2026-06-21-lead-context.md.claude/skills/agent-build-propose/proposal-template.md
Read additional docs when the proposed change touches their authority surface. Do not summarize orientation docs in the proposal unless they change the recommendation.
Output
- If the user gives an existing packet path, update that packet's
proposal.md. - Otherwise create
experiments/01-agentic-build/changes/YYYY-MM-DD-short-slug/proposal.md. - Choose a short lowercase hyphen slug from the requested change.
- Use
proposal-template.mdas the structure; do not add new sections unless the proposal truly needs them.
Rules
- Only create or update the packet
proposal.md; do not make implementation changes. proposal.mdis a plan-only artifact — the thingproposal-mode review judges and the lead decides before any code. Keep implementation evidence out of it: status never flips toimplemented, and receipts / gate logs land inimplementation-notes.mdafter the decision. Don't let the proposal and the change collapse into one file — that boundary is what lets review gate instead of just narrate.- Keep frame context inside
proposal.md; if the problem is too unclear for that, stop and say a separate frame is needed. - Fill the required fields:
risk,write_scope,spend_stop,acceptance_checks,escalate_if. - If a required field is unknown, write
unknownplus why. - Diverge broadly in scratch, then compare only the strongest 2-3 finalist options before recommending.
- Prefer the simplest option that can meet the acceptance checks this proposal defines.
- Follow the size defaults in
2026-06-21-lead-context.md. - Treat
highrisk as planning-only unless Brian explicitly approves implementation later. - Log an observation only for a reusable miss, Brian correction, severe risk, or recurring process friction. Routine work-phase judgment stays private.
Final Response
End with:
- proposal path, or
no proposal writtenplus why - one-sentence recommendation
- risk, if a proposal was written
- review needed: approve, revise, park, or cut before implementation
No implementation changes were made.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

