[aliperdehan]

Second slice of roadmap 1a: peak editing in irspec edit, on manifest.json only.

irspec edit DIR FILE peak add X [X ...]      # each X snapped to the nearest local band
irspec edit DIR FILE peak remove X [X ...]   # removes the stored peak nearest X within 15 cm-1
irspec edit DIR FILE peak clear
irspec edit DIR FILE peak auto               # re-pick with the entry's threshold/prominence/distance

How it works

• The peaks are worked out by session's own code (ir_organize._finish / resolve_peaks, hence find_peaks_manual, find_peaks_auto, assign_label_y), so snapping, label heights and rounding are exactly what session produces. peak auto on a fresh session changes nothing, byte for byte.
• add, remove and clear write an explicit peaks list, which session keeps. They start from what the figure shows now, so an entry with no peaks key starts from the automatic list.
• The spectrum is read as session reads it: an own .SPA from the file (this is what makes auto match session; the plain CSV's 3-decimal x flips a tie, 1446.3 against 1446.4). If the .SPA is gone, the copies in plain/ or csv/ are used. A reference spectrum is read from plain/NAME.csv. If none exists, exit 2 on stderr.
• Same conventions as set/unset: atomic write, entry and key order kept, idempotent, each change printed as old -> new, --json prints exactly one object. Exit 2 means the command cannot be carried out; exit 1 means the manifest is missing or broken.
• remove finds nothing within the tolerance: no change, a note on stderr, exit 0.
• peak auto says it replaces hand edits and lists what it dropped.

One change outside edit.py

ir_organize.resolve_peaks treated "peaks": [] like a missing key and picked again, so peak clear (or removing the last peak) would have been undone by the next irspec session. It now takes [] as "no peaks"; deleting the key still picks again. The README, CHANGELOG and roadmap say so. A legacy empty string for peaks now also means no peaks instead of picking.

Tests

tests/test_edit_peaks.py has 31 tests on the sample .SPA and the phenacetin_T.jdx fixture. One runs irspec session --build none after edits and checks that they survive and reach figures.json and the .md. Another does the same after clear. Each test was mutation-checked: 17 mutants of edit.py and ir_organize.py, all killed. One survived at first (adding a band that is already present), and a test was added for it.

Full suite on Python 3.13 in a venv: 120 tests, OK, 19 skipped. lualatex and R are not installed in this environment, so the TeX and R paths were not run. Please run the suite locally before merging.

Not touched: version, tags, session.py, irspec-ir.sty, rename/remove/move, per-figure limits, --no-build, preview. Not merged.

🤖 Generated with Claude Code

https://claude.ai/code/session_01GqBTMN8XHvB3aG8T6PmfYS