qual-coding
Skill0 starsChanged 49 days agoMITView on GitHub
Qualitative coding with an auditable codebook — open, axial and selective coding on interview or transcript data, with anchored quotes and memo trail. Use for 質性編碼, thematic analysis, grounded theory coding, interview analysis, or when the user needs to code transcripts and build a codebook.
In this repository2 other files it gives its agents.
- emi-transcriptSkill
- lit-reviewSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 3 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 18
If the transcript is machine-generated, flag it: ASR errors are not data.
- Distinctive: line 47
Every category gets a memo: what it is, why it matters, what contradicts it, what to look for next.
- Distinctive: line 68
Never invent illustrative quotes.
- Distinctive: line 70
Do not claim saturation unless you have coded to the point of no new codes and can show when that happened.
≈ “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
Qualitative coding
Turns raw transcripts into a codebook where every code is anchored in real text.
references/evidence-contract.md is binding, with one addition: for qualitative
work the anchor quote is the source of truth, not the code name.
Step 0 — Prepare
- Normalise the transcript: one utterance per row, keep speaker labels and timestamps.
- Pseudonymise participants before anything else (
P01,T01). Never write real names, schools or places into outputs. - Save as
data/transcript.tsv:id \t speaker \t timestamp \t text
If the transcript is machine-generated, flag it: ASR errors are not data. Mark
uncertain segments [inaudible] and exclude them from coding.
Step 1 — Open coding
Work line by line. For each segment either assign an existing code or create one.
A new code needs: code, definition, inclusion, exclusion, example_id.
Codes must be close to the data in this pass. Do not import a framework yet.
Step 2 — Axial coding
Group codes into categories, and look for:
- conditions → action/interaction → consequences
- recurring sequences
- negative cases (data that contradicts the emerging pattern)
Negative cases are reported, not discarded.
Step 3 — Selective coding
Name the core category and state, in one paragraph, how the other categories relate to it. If no core category is defensible, say so — a review with no single core category is a valid result.
Step 4 — Memos
Every category gets a memo: what it is, why it matters, what contradicts it, what to look for next. Memos are dated and never deleted.
Outputs
codebook.md:
### [code name]
- definition:
- inclusion / exclusion:
- segments: n
- anchors: [S012] "..." · [S078] "..."
- memo:coding-matrix.csv: segment_id, speaker, timestamp, codes, memo_id
Hard constraints
- No code without at least one anchor quote. An unanchored theme is a fabrication.
- Never invent illustrative quotes. If you need an example, pull a real one or leave it blank.
- Report code frequency as counts of segments, not as percentages of "participants who said" unless you actually counted participants.
- Do not claim saturation unless you have coded to the point of no new codes and can show when that happened.
- Inter-coder agreement: if a second coder exists, report the procedure and disagreements. If not, state that coding was single-author.
Files
data/transcript.tsv
codebook.md
coding-matrix.csv
memos/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.

