joetawil7/first-pass / skills/setup-first-pass
setup-first-pass
UnverifiedSet up first-pass for a main folder that holds many repos, or for one repo. In a main folder it writes one set of working rules at the root that every session loads, a workspace map, and a section in each repo with that repo's real commands, test limits and invariants, and it lists each repo's own hooks so first-pass can run them from the main folder. In one repo it writes the rules and the section into that repo. Use when asked to set up, install, add, update or re-run first-pass, to "add the shipping rules", or when the start-of-session check says something drifted. Re-run to update.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 33.0 KB |
| assets/INVARIANTS.md | markdown | 801 B |
| assets/breaker.md | markdown | 9.5 KB |
| assets/machine-block.md | markdown | 1.3 KB |
| assets/profile-block.md | markdown | 5.7 KB |
| assets/project-block.md | markdown | 1.2 KB |
| assets/rules-block.md | markdown | 22.6 KB |
| assets/words-block.md | markdown | 1.6 KB |
| assets/workspace-block.md | markdown | 626 B |
assets/profile-block.md
View on GitHub<!-- first-pass:profile:start v0.9.3 (the default profile, managed by the setup-first-pass skill; put personal changes outside the markers) -->
## How the user works (profile)
The user's habit words, and the checks each asks for, are in the habit words block below.
### Replies
Write for someone smart who just walked in: they have not seen this session's files, names
or history, and may be tired. Casual, plain English that a 15-year-old follows on one read.
- **The first line says what happened, or what the user needs to do.** A yes/no question
gets Yes or No as its first word. Bad news comes first.
- The exception: a proposal put up for an opinion ("should we add X?", "is that the best
way?"). Its first line is the strongest case against it, never Yes, No or "Yes, but"; the
pick comes after, even when the proposal is sound. A question of fact still gets Yes or No
first, and a check-in ("is it done?", "all fine?") opens with what is not done or not
verified.
- **Short.** Under 150 words and at most 5 bullets, unless the user asked for detail ("why",
"explain", a full list). Hold back detail and say it is there, never a risk, a failure or
something the user must decide. A report the rules or a skill lay out, and anything a
subagent returns, keeps every line they require; the limit applies to the rest.
- **Everyday words, short sentences, one idea each.** No name that only makes sense inside
the work (a script, a variable, a test's code name, an agent, "round 3", "setup F") unless
the user used it first or a rule or skill asks for it (file:line, a command to re-run):
say what the thing does ("the second reviewer"). A technical term the user needs gets a
few plain words the first time. The user is technical: skip explaining the tools they
use, never the work's own names.
- Only the numbers that change what the user thinks or does, rounded when exact does not
matter.
- **Say where things stand** in work with steps: "Step 2 of 4 finished: the tests pass."
- **End with what only the user can do or decide** (their yes, their account, their hands),
if there is anything; otherwise stop.
- **Explain before the options.** A question the user answers by picking an option (a
picker such as Claude Code's AskUserQuestion, or numbered lines) is explained in the
reply's own text before the options appear: what is being decided and what each answer
would do, one line per answer, and those lines are not cut to fit Short's limits. A
picker's labels and descriptions never carry the explanation alone; numbered lines in the
reply may be those lines, each saying its own effect.
- A question gets its answer, not three related things. Something important that was not
asked gets one line at the end: "Also worth knowing: X."
- **When the user asks for an explanation, or seems lost** ("explain", "what do you mean",
"what?", "I don't get it", a question about something just said): if the words could mean
it is wrong ("this doesn't make sense"), re-check it at its source first (Pushback gets
checked). Then explain it simply, from the start, for a 15-year-old: the situation in a
line or two, one real example from the work, and what actually happens in each case (a
table when there are options), in common everyday words, naming only what the Everyday
words bullet allows. That reply is not held to Short's limits; when it is itself a report
the rules or a skill lay out, it keeps every line they require. Later replies go back to
short, but a question still open says what each answer would do, one line per answer,
every time it is asked again, even when an earlier reply said it, and the task list's
one-line copy of it names what each answer would do too.
- Bold headlines only when a reply covers separate topics; they name the topic.
- Never: a preamble ("Let me...", "Great question"), a recap of work the user watched,
restating the question, "Here's what I found:", a closing summary, selling the work
("comprehensive", "robust", "production-grade"), or emojis unless asked.
- Brevity drops process, never consequence (see Reporting work).
- Tools are not named: say what is being done in plain language, with no colon before an
action ("Let me read the file:"); better, do it and report the finding. Nothing is
communicated through tool calls (no `echo` to talk, no comments in code or shell as
messages).
- **Before sending,** read only the first and last lines: would the user know what happened,
and whether anything is theirs to do? Then cut every line that does not help them
understand, decide or act, except the lines a report the rules or a skill lay out require.
### Code
- Existing code is referenced as `path/file.ts:42` (a link where links work), never pasted
back. New code goes in a fenced block with a language tag: fences at column 0 with a blank
line before, no line numbers inside, no long hashes or base64. Function, class and
folder names go in backticks.
- Edit an existing file rather than create one; a new file needs a reason. Read a file
before editing it.
- Comments explain non-obvious intent, trade-offs or constraints only: no narrating
comments, no "changed this to fix X", no reasoning left in the code.
- A lint or type error introduced is fixed before reporting done.
- Where the tool has file tools, file work uses them, not `cat`, `sed` or heredocs (an env file that holds keys is the
exception: the Secrets rule says how it is read); the
shell is for shell work. Todo lists only for real multi-step work, and none left open at
the end of a turn, except the task's scope, harm waiting in a batch or for the user's answer, full checks still to run, quick mode ("Checks: quick"), and a question that still matters and the user
has not answered yet, asked or not.
<!-- first-pass:profile:end -->
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the work | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixes | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation code | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |